KR102600837B1 - Accident Prevention System Using Deep Learning - Google Patents

Abstract

The embodiment relates to an accident prevention system using deep learning.
Specifically, this system synchronizes the time between the tag of the smart gateway and the master gateway RTLS server, and sequentially performs communication between the tag and anchor through time division.
That is, all tags are set to the same time by synchronizing the time of tags applied to multiple moving objects with the RTLS server of the master gateway, and each tag sequentially performs 1:1 communication with the anchor by time division.
Therefore, after synchronizing the time between the tag of the smart gateway and the RTLS server of the master gateway, communication between the tag and the anchor is carried out sequentially through time division, so that accurate distance measurement and location and Speed can be measured.
And, through this, the accuracy of predicting collisions and dangerous situations is improved. In addition, because the time of each moving object is synchronized, each moving object stores images for the same time, and it is easy to analyze the images for the same time for each moving object when a collision or dangerous situation occurs.

Description

Accident Prevention System Using Deep Learning}

The content disclosed in this specification relates to an accident prevention system using deep learning, and more specifically, to an accident prevention system using deep learning that predicts the possibility of collision between construction machines such as forklifts and warns of danger when there is a possibility of collision. will be.

Unless otherwise indicated herein, the material described in this section is not prior art to the claims of this application, and is not admitted to be prior art by inclusion in this section.

Conventional devices or judgment systems that determine the possibility of collision between vehicles use ultrasonic sensors, video cameras, and other sensors to find the distance between the vehicle and surrounding vehicles, and if the distance is within a certain range, warn of the risk of collision or apply the brakes. The main goal is to prevent collisions or reduce damage.

These methods require multiple sensors to be installed in each vehicle and a device to process signals exchanged between multiple sensors. Since multiple sensors and devices are used, power consumption and costs may increase and the system failure rate may increase.

Additionally, as an example of prior art, Republic of Korea Patent No. 10-0947559 proposes a configuration that detects an object by using at least two RF oscillators to alternately oscillate multiple frequencies according to a certain period. .

However, in the case of the prior art, there is a problem that a dead point phenomenon occurs periodically in which an object cannot be detected even when it approaches, and there is a problem that the detection distance is narrow.

KR 100947559 B1 KR 102004518 B1 KR 1020190114327 A On the other hand, especially in this regard, the conventional technology mounts a UWB module for operating a tag function on a moving object and applies a UWB module for operating an anchor function in a certain area. So, the distance from each anchor is calculated from the tag and transmitted to the relay gateway, and the relay gateway transmits it to the master gateway, and the position and speed of each tag are calculated by the master gateway's RTLS algorithm. It is a structure that predicts situations about collisions and risks through . However, at this time, the tag and anchor communicate 1:1, and if time synchronization is not performed, multiple tags will attempt to communicate with the anchor, and only one tag will communicate with the anchor. Since communication is limited to , the remaining tags cannot calculate the distance from the anchor. This causes the accuracy of prediction of collisions and risks to decrease due to the inability to obtain location information for the moving object at the same time, which can lead to accidents. For reference, the prior literature on this technology is as follows. KR 102308601 Y1 Document 1 relates to a method for positioning a moving object, and is a technology for time synchronizing a plurality of slave anchors with a master anchor and a clock.

The disclosed content synchronizes the RTLS server time of the tags applied to multiple moving objects and the master gateway to set all tags to the same time, and each tag sequentially performs 1:1 communication with the anchor to prevent communication errors and delays between tags and anchors. We aim to provide an accident prevention system using deep learning to eliminate .

The accident prevention system using deep learning according to the embodiment is,

In detail, it consists of a smart gateway, relay gateway, and master gateway.

Additionally, multiple smart gateways and multiple relay gateways each communicate with the master gateway through a mesh network configuration.

In this case, the smart gateway is composed of a short-range communication mesh, that is, LoRa Mesh, and the relay gateway is composed of Wi-Fi Mesh.

In addition, time synchronization synchronizes the time by transmitting the RTLS server time within the master gateway to the LoRa of the smart gateway through LoRa mounted on the master gateway.

And, at this time, in the case of a mobile entity located in a communication shadow area, it is transmitted through the LoRa Mesh network.

So, each time-synchronized tag sequentially communicates with the anchor to calculate the distance, and the calculated distance is delivered to the anchor through UWB communication, and the anchor transmits the ID and distance information of each tag to the Wi-Fi mesh (Wi-Fi mesh). It is transmitted to the master gateway through the Fi Mesh network.

In addition, through this, the master gateway analyzes the distance information of each received tag from the RTLS server and, when a collision or dangerous situation is predicted, transmits it to the moving object of the tag through the master gateway's roller to prevent accidents. do.

According to embodiments, after time synchronization of the tag of the smart gateway and the RTLS server of the master gateway, communication between the tag and the anchor is performed sequentially through time division, thereby accurately measuring the distance without delay or error in communication between the tag/anchor. Position and speed can be measured.

And, through this, the accuracy of predicting collisions and dangerous situations is improved. In addition, because the time of each moving object is synchronized, each moving object stores images for the same time, and it is easy to analyze the images for the same time for each moving object when a collision or dangerous situation occurs.

1 is a conceptual diagram schematically showing an accident prevention system using deep learning according to an embodiment.
Figure 2 is a diagram showing the overall accident prevention system using deep learning according to an embodiment.
Figure 3 is a block diagram briefly showing the configuration of an edge terminal, a slave gateway, and a master gateway applied to an accident prevention system using deep learning according to an embodiment.
Figure 4 is a procedure flowchart showing the operation of an accident prevention system using deep learning according to an embodiment.
FIGS. 5A and 5B are diagrams for explaining in detail the time synchronization operation applied to the accident prevention system using deep learning according to an embodiment.

Figure 1 is a diagram for conceptually explaining an accident prevention system using deep learning according to an embodiment.

As shown in FIG. 1, the accident prevention system 1000 using deep learning in one embodiment basically includes a mobile device, that is, an edge terminal 100, a slave gateway 200 (including a smart gateway and a relay gateway), and a master gateway ( 300) and control server 400.

In this case, according to one embodiment, the system 1000 synchronizes the time between the tag of the smart gateway 200 and the RTLS server of the master gateway 300, and then synchronizes the time between the tag and the anchor. Communication is performed sequentially through time division.

That is, all tags are set to the same time by synchronizing the tags applied to the multiple moving objects 100 and the RTLS server of the master gateway 300, and each tag sequentially performs 1:1 communication with the anchor by time division.

Therefore, this eliminates communication delays and errors between tags/anchors and enables accurate distance measurement and location and speed measurement.

Specifically, this system 1000 is as follows.

That is, first, a tag-equipped moving object (edge terminal 100) including construction machinery passing within the site is basically applied.

In this case, a smart gateway is installed in a number of different locations on the site and collects the tag ID and distance data obtained from tags for a number of different moving objects (edge terminal 100) for each moving object. Includes (200).

In addition, it also includes a relay gateway 200 that is installed in a plurality of different sectors in the field according to the same data transmission range and relays the tag ID and distance data for each corresponding moving object.

Therefore, a master gateway 300 is also provided that receives the tag ID and distance data for each moving object in batches from the relay gateway 200, calculates the position and speed for each moving object, and provides information on the risk of collision with surrounding moving objects. .

In this state, according to one embodiment, this system 1000 performs the following services.

a) First, for this purpose, the smart gateway 200 is configured as a short-range communication mesh, and the relay gateway 200 is configured as a Wi-Fi mesh.

b) So, the master gateway 300 synchronizes time with each smart gateway 200 through the short-range communication mesh and the Wi-Fi mesh by broadcasting the internal RTLS server time, and also new mobile devices at each location. Each time it detects, time synchronization is performed repeatedly by broadcasting.

c) In the case of such time synchronization, a signal is transmitted from the tag for each moving object to the anchor, and distance data for each tag ID is transmitted from the master gateway 300 to the Wi-Fi mesh through the anchor.

d) Therefore, through this, the master gateway 300 analyzes the distance data for each tag ID in the RTLS server, calculates the position and speed for each moving object, and provides information on the risk of collision with surrounding moving objects.

Therefore, through this, the system according to one embodiment synchronizes the time between the tag of the smart gateway and the RTLS server of the master gateway, and then communicates between the tag and the anchor sequentially through time division, thereby ensuring accurate communication without delay or error between tags/anchors. It is possible to measure distance and thereby measure position and speed.

And, through this, the accuracy of predicting collisions and dangerous situations is improved. In addition, because the time of each moving object is synchronized, each moving object stores images for the same time, and it is easy to analyze the images for the same time for each moving object when a collision or dangerous situation occurs.

Figure 2 is a diagram showing the overall accident prevention system using deep learning according to an embodiment.

As shown in FIG. 2, an accident prevention system using deep learning according to an embodiment includes a plurality of different mobile objects (tags) 100 and smart gateways 200 passing within various field locations, and each sector grouping multiple locations. Includes the installed relay gateway 210 and master gateway 220.

Additionally, this system includes a management information processing device 300, a troubleshooting information processing device 400-1 that provides additional services in connection with the outside, and a fire department information processing device 400-2, etc. Devices are connected to each other through their own network, for example, a TCP/IP network or an LTE network.

The mobile object 100 (i.e. tag) travels through a number of different on-site locations and may be composed of an edge terminal equipped with a tag, and is installed on construction equipment such as forklifts. For example, this mobile object 100 includes a UWB tag module, LoRa T module, WiFi module, shooting module, display module, alarm module, and information processing module.

The smart gateway 200 is installed in a number of different locations on the site and collects tag IDs and distance data obtained from tags for a number of different moving objects (edge terminals 100) for each moving object.

The relay gateway 210 is installed in a plurality of different sectors within the field according to the same data transmission range, and relays the tag ID and distance data for each corresponding moving object. That is, data is relayed between the smart gateway 200 and the master gateway 220.

The master gateway 220 receives the tag ID and distance data for each moving object in batches from the relay gateway 210, calculates the position and speed for each moving object, and provides information on the risk of collision with surrounding moving objects. And, according to one embodiment, the master gateway 220 synchronizes the tags applied to the plurality of mobile objects 100 with the internal RTLS server time to set all tags to the same time, and each tag performs 1:1 communication with the anchor. It is performed sequentially by time division. Therefore, this eliminates communication delays and errors between tags/anchors and enables accurate distance measurement and location and speed measurement.

Figure 3 is a block diagram showing the configuration of a moving object applied to an accident prevention system using deep learning according to an embodiment, that is, an edge terminal 100, a slave gateway 200, and a master gateway 300.

As shown in FIG. 3, the mobile object 100 according to one embodiment largely includes a UWB tag module 110, a LoRa T module 120, a WiFi module 130, a photographing module 140, and a display module ( 150), an alarm module 160, and an information processing module 170.

In addition, the slave gateway 200 according to one embodiment exchanges signals with the edge terminal 100 and acquires location data of the edge terminal 100, using the UWB anchor module 210 and the LoRa T module 220. ), and is composed of a control module 230.

In addition, the master gateway 300 according to one embodiment receives data from the slave gateway 200 and detects collisions between the edge terminals 100, including the WiFi module 310 and the LoRa G module 320. , it is composed of a computation module (not shown), an Ethernet module (330), and a control module (340).

<Accident prevention system using deep learning according to the first embodiment of the present disclosure>

First, the accident prevention system using deep learning according to the first embodiment of the present disclosure may be configured to include an edge terminal 100, a slave gateway 200, a master gateway 300, and a control server 400. .

The edge terminal 100 is provided on construction equipment such as forklifts and includes a UWB tag module 110, a LoRa T module 120, a WiFi module 130, a photography module 140, a display module 150, It may be configured to include an alarm module 160 and an information processing module 170.

The UWB tag module 110 is a positioning sensor using UWB (Ultra Wide Band), which is a wireless communication technology that uses a radio with a very wide bandwidth. This UWB method uses impulse wireless signals, and thus, unlike conventional narrow-band radio communication, it has the advantage of simplifying the structure by not using a carrier wave. Additionally, the UWB method has the advantage of providing high temporal resolution, wide bandwidth, and being less affected by multipath and interference compared to existing radios.

The UWB tag module 110 broadcasts its location to the UWB anchor module 210 of the slave gateway 200, which will be described later, and the received UWB anchor module 210 is sent to the UWB tag module 110. Gives a response.

Specifically, the UWB tag module 110 may be configured to transmit a tag ID, broadcasting time, response time, and current transmission time to the UWB anchor module 210.

The photographing module 140 may be configured to capture an image in a predetermined direction and implement the captured image in the display module 150.

Meanwhile, the information processing module 170 detects a predetermined object based on image data captured by the photographing module 140, and can be configured to generate an event when the predetermined object approaches within a predetermined radius. .

Specifically, the information processing module 170 detects an object based on the image captured by the photographing module 140, determines whether the detected object is a predetermined object, and then determines whether the determined object is within a predetermined range. When approaching, an event may be generated and a signal may be sent to the alarm module 160 to provide a warning alarm.

To explain in more detail, the information processing module 170 samples the image captured by the photographing module 140 at a predetermined time and interval interval, and calculates the pixel difference between the images based on the sampled image. It can be configured to detect an object using .

The information processing module 170 may be configured to extract features from the detected object and determine whether the detected object is a predetermined object by a multiple artificial neural network that has been deep-learned in advance based on the extracted features. there is.

This multiple artificial neural network is deep-learned with data on the characteristics of multiple objects in advance, and includes an input layer, four or less hidden layers, and an output layer, and is calculated to minimize the root mean square deviation between the input layer and the output layer. and can be configured to be controlled.

The preset object may include a person or a construction machine such as a forklift, but is not limited thereto and may be configured to be configured in various settings as needed.

Meanwhile, binarization techniques and morphological techniques can be used to extract features from the detected object, and of course, various methods can be applied as needed.

The alarm module 160 activates an alarm when it receives a predetermined signal, and may consist of an auditory alarm such as a buzzer as well as a visual alarm such as an LED.

The LoRa T module 120 may be configured to transmit event information generated by the information processing module 170 to the master gateway 300, which will be described later, and to receive collision information from the master gateway 300. This LoRa (Long Range) module method has the advantage of enabling long-distance communication with low power.

The WiFi module 130 may be configured to transmit image data captured by the photographing module 140 to the master gateway 300, which will be described later.

Meanwhile, the edge terminal 100 may further include an acceleration sensor module (not shown), and the acceleration sensor module (not shown) may be configured to be used as an auxiliary device for measuring the impact and speed transmitted during a collision. You can. This will allow more precise speed measurement and collision information to be secured.

The slave gateway 200 exchanges signals with the edge terminal 100 and acquires location data of the edge terminal 100, using the UWB anchor module 210, LoRa T module 220, and control module 230. It may be configured to include. In this case, according to one embodiment, the slave gateway 200 includes a smart gateway and a relay gateway.

The UWB anchor module 210 acquires location data of the edge terminal 100 by exchanging signals with the UWB tag module 110 of the edge terminal 100 described above.

The acquired data is transmitted by the LoRa T module 220 to the master gateway 300, which will be described later.

The UWB anchor module 210 and the LoRa T module 220 may be controlled through the control module 230. For example, according to one embodiment, when smart gateways and relay gateways are installed in a plurality of different locations on the site, the control module 230 tags each of a plurality of different moving objects (edge terminal 100). The tag ID and distance data obtained from are collected for each moving object.

It is preferable that three or more slave gateways 200 are installed in order to secure accurate location data, but this is not limited, and of course, they can be added or subtracted as needed.

The master gateway 300 receives data from the slave gateway 200 and detects collisions between the edge terminals 100, including the WiFi module 310, LoRa G module 320, and calculation module (not shown). , it can be configured to include an Ethernet module 330 and a control module 340. And, according to one embodiment, the master gateway 300 synchronizes the tags applied to the plurality of mobile objects 100 with the internal RTLS server time to set all tags to the same time, and each tag performs 1:1 communication with the anchor. It is performed sequentially by time division. Therefore, this eliminates communication delays and errors between tags/anchors and enables accurate distance measurement and location and speed measurement.

The master gateway 300 receives the image captured by the photographing module 140 of the edge terminal 100 through the WiFi module 310.

The calculation module (not shown) calculates the speed of the edge terminal 100 based on the data transmitted from the slave gateway 200, and calculates the speed of the edge terminal 100 between the edge terminal 100 and another edge terminal 100 nearby. Calculate the distance. The tag ID and distance data for each moving object are collectively transmitted from the relay gateway 210, the position and speed of each moving object are calculated, and collision risk information with surrounding moving objects is provided. And, according to one embodiment, this operation module synchronizes the tags applied to the plurality of moving objects 100 with the internal RTLS server time to set all tags to the same time, and each tag performs 1:1 communication with the anchor through time sharing. Perform sequentially. Therefore, this eliminates communication delays and errors between tags/anchors and enables accurate distance measurement and location and speed measurement.

Additionally, the calculation module (not shown) may be configured to detect the risk of collision of the edge terminal 100 based on the calculated data. Specifically, if another edge terminal 100 exists within a predetermined distance according to the speed of the edge terminal 100, it may be configured to determine that there is a risk of collision. Meanwhile, the predetermined distance may be configured to change depending on the speed of the edge terminal 100.

When the calculation module (not shown) detects a collision risk of the edge terminal 100, it can be configured to transmit the collision risk information to the edge terminal 100 through the LoRa G module 320. In addition, the edge terminal 100, which has received the collision risk information, may be configured to provide a collision warning alarm by the warning module 160.

Meanwhile, the master gateway 300 may be configured to further include an Ethernet module 330, and the master gateway 300 may be configured to operate a control server (to be described later) through the Ethernet module 340 or the WiFi module 310. 400) may be configured to transmit data.

Specifically, when the calculation module (not shown) detects a collision risk of the edge terminal 100, the location, speed, and collision risk information of the edge terminal 100 are transmitted to the control server 400, which will be described later. , can be configured to be monitored by the control server 400.

The control module 340 may be configured to control the WiFi module 310, LoRa G module 320, Ethernet module 330, and calculation module (not shown).

The control server 400 can receive data from the master gateway 300, convert it into a database, and monitor it. It can be configured in the form of a general server.

The control server 400 includes image data captured by the photographing module 140 of the edge terminal 100, event information generated by the information processing module 170 of the edge terminal 100, and the slave gateway 200. It may be configured to receive the location data transmitted from, the speed data calculated by the master gateway 300, and the collision risk information detected by the master gateway 300, convert them into a database, and monitor them.

Specifically, based on the data, the map may be configured to display the location of the edge terminal 100 and the object detected by the edge terminal 100, and the speed of the edge terminal 100 and the edge by GUI. It can be configured to provide distance between terminals 100, collision risk information of the edge terminal 100, and be monitored through video record inquiry of the edge terminal 100, etc.

<Accident prevention system using deep learning according to the second embodiment of the present disclosure>

Next, the accident prevention system using deep learning according to the second embodiment of the present disclosure may be configured to include an edge terminal 10, a slave gateway 200, a master gateway 300, and a control server 400. (For reference, the configuration according to one embodiment is the same as the configuration of the first embodiment above).

The edge terminal 100 is provided on construction equipment such as forklifts, and includes a UWB tag module 110, LoRa T module 120, WiFi module 130, photography module 140, display module 150, and alarm. It may be configured to include a module 160 and an information processing module 170.

The UWB tag module 110 is a positioning sensor using UWB (Ultra Wide Band), which is a wireless communication technology that uses a radio with a very wide bandwidth. This UWB method uses impulse wireless signals, and thus, unlike conventional narrow-band radio communication, it has the advantage of being simple in structure by not using a carrier wave. Additionally, the UWB method has the advantage of providing high temporal resolution, wide bandwidth, and being less affected by multipath and interference compared to existing radios.

The UWB tag module 110 broadcasts its location to the UWB anchor module 210 of the slave gateway 200, which will be described later, and the received UWB anchor module 210 is sent to the UWB tag module 110. Gives a response.

Specifically, the UWB tag module 110 may be configured to transmit a tag ID, broadcasting time, response time, and current transmission time to the UWB anchor module 210.

The photographing module 140 may be configured to capture an image in a predetermined direction and implement the captured image in the display module 150.

Meanwhile, the information processing module 170 detects a predetermined object based on the image data captured by the photographing module 140, and can be configured to generate an event when the predetermined object approaches within a predetermined radius. there is.

Specifically, the information processing module 170 detects an object based on the image captured by the photographing module 140, determines whether the detected object is a predetermined object, and then determines whether the determined object is within a predetermined range. When approaching inside, an event may be generated and a signal may be sent to the alarm module 160 to provide a warning alarm.

To explain in more detail, the information processing module 170 samples the image captured by the photographing module 140 at a predetermined time and interval interval, and calculates the pixel difference between the images based on the sampled image. It can be configured to detect an object using .

The information processing module 170 may be configured to extract features from the detected object and determine whether the detected object is a predetermined object by a multiple artificial neural network that has been deep-learned in advance based on the extracted features. there is.

This multiple artificial neural network is deep-learned with data on the characteristics of multiple objects in advance, and includes an input layer, four or less hidden layers, and an output layer, and is calculated to minimize the root mean square deviation between the input layer and the output layer. and can be configured to be controlled.

The preset object may include a person or a construction machine such as a forklift, but is not limited thereto and may be configured to be configured in various settings as needed.

Meanwhile, binarization techniques and morphological techniques can be used to extract features from the detected object, and of course, various methods can be applied as needed.

The alarm module 160 activates an alarm when it receives a predetermined signal, and may consist of an auditory alarm such as a buzzer as well as a visual alarm such as an LED.

The LoRa T module 120 may be configured to transmit event information generated by the information processing module 170 to the master gateway 300, which will be described later, and to receive collision information from the master gateway 300. This LoRa (Long Range) module method has the advantage of enabling long-distance communication with low power.

The WiFi module 130 may be configured to transmit image data captured by the photographing module 140 to the master gateway 300, which will be described later.

Meanwhile, the edge terminal 100 may further include an acceleration sensor module (not shown), and the acceleration sensor module (not shown) may be configured to be used as an auxiliary device for measuring the impact and speed transmitted during a collision. You can. This will allow more precise speed measurement and collision information to be secured.

The slave gateway 200 exchanges signals with the edge terminal 100 and acquires location data of the edge terminal 100, using the UWB anchor module 210, LoRa T module 220, and control module 230. It may be configured to include.

The UWB anchor module 210 acquires location data of the edge terminal 100 by exchanging signals with the UWB tag module 110 of the edge terminal 100 described above.

The acquired data is transmitted by the LoRa T module 220 to the master gateway 300, which will be described later.

The UWB anchor module 210 and the LoRa T module 220 may be controlled through the control module 230.

It is preferable that three or more slave gateways 200 are installed in order to secure accurate location data, but this is not limited, and of course, they can be added or subtracted as needed.

The master gateway 300 receives data from the slave gateway 200 and transmits it to the control server 400, which will be described later, and includes a WiFi module 310, LoRa G module 320, Ethernet module 330, and control. It may be configured to include a module 340.

The master gateway 300 receives the image captured by the photographing module 140 of the edge terminal 100 through the WiFi module 310.

The master gateway 300 may be configured to receive location data of the edge terminal 100 from the slave gateway 200 through the LoRa T module 320.

The data transmitted from the edge terminal 100 and the slave gateway 200 may be configured to be transmitted to the control server 400, which will be described later, through the Ethernet module 330 or the WiFi module 310.

In addition, the master gateway 300 may be configured to receive collision risk information from the control server 400, which will be described later, and transmit the collision risk information to the edge terminal 100, and the collision risk information received The edge terminal 100 may be configured to provide a collision warning alarm by the warning module 160.

The control server 400 can receive data from the master gateway 300, convert it into a database, and monitor it. It can be configured in the form of a general server.

The control server 400 receives data from the master gateway 300, calculates the speed of the edge terminal 100, and calculates the distance between the edge terminal 100 and another edge terminal 100 nearby. Calculate .

Additionally, the control server 400 may be configured to detect the risk of collision of the edge terminal 100 based on the calculated data. Specifically, if another edge terminal 100 exists within a predetermined distance according to the speed of the edge terminal 100, it may be configured to determine that there is a risk of collision. Meanwhile, the predetermined distance may be configured to change depending on the speed of the edge terminal 100.

When the control server 400 detects a collision risk of the edge terminal 100, it may be configured to transmit the collision risk information to the above-described master gateway 300, and the master gateway 300 Collision risk information may be transmitted to the edge terminal 100, and the edge terminal 100 may be configured to provide a collision warning alarm by the warning module 160.

Additionally, the control server 400 may be configured to store and monitor data received from the master gateway 300 and data calculated by the control server 400 into a database.

The control server 400 includes image data captured by the photographing module 140 of the edge terminal 100, event information generated by the information processing module 170 of the edge terminal 100, and the slave gateway 200. It can be configured to monitor the location data transmitted from ), the speed data calculated by the control server 400, and the collision risk information detected by the control server 400 into a database.

Specifically, based on the data, the map may be configured to display the location of the edge terminal 100 and the object detected by the edge terminal 100, and the speed of the edge terminal 100 and the edge by GUI. It can be configured to provide distance between terminals 100 and collision risk information of the edge terminal 100, and to be monitored through video record inquiry of the edge terminal 100.

Accordingly, the accident prevention system using deep learning according to an embodiment of the present disclosure,

Using the UWB module, object collisions can be detected based on stable and high-precision location data. The structure can be simplified using the UWB module, and it provides a wider bandwidth and higher temporal resolution than the existing radio method. It has the advantage of doing so.

In addition, power efficiency can be increased using the LoRa module, it can be applied in various environments due to excellent coverage, can be easily installed and expanded, and has the advantage of minimizing the impact of interference.

In particular, after synchronizing the time between the smart gateway's tag and the master gateway's RTLS server, communication between the tag and anchor is carried out sequentially through time division, so accurate distance measurement and position and speed measurement without communication delay or error between tags/anchors This is possible.

And, through this, the accuracy of predicting collisions and dangerous situations is improved. In addition, because the time of each moving object is synchronized, each moving object stores images for the same time, and it is easy to analyze the images for the same time for each moving object when a collision or dangerous situation occurs.

Figure 4 is a procedural flowchart sequentially showing the operation of an accident prevention system using deep learning according to an embodiment.

As shown in FIG. 4, the accident prevention system using deep learning according to one embodiment is first composed of a) the smart gateway and a short-range communication mesh in the master gateway, and the relay gateway and the It consists of a pie mesh.

b) Next, the master gateway performs time synchronization with each gateway by broadcasting the internal RTLS server time through the short-range communication mesh and the Wi-Fi mesh, and also detects a new moving object while performing time synchronization. Each time, time synchronization is performed repeatedly by re-broadcasting.

c) So, when such time synchronization is performed, the signal from the tag for each moving object is transmitted as distance data for each tag ID to the Wi-Fi mesh through the anchor.

d) Through this, the RTLS server analyzes the distance data for each tag ID, calculates the position and speed for each moving object, and provides information on the risk of collision with surrounding moving objects.

Therefore, through this, the system according to one embodiment synchronizes the time between the tag of the smart gateway and the RTLS server of the master gateway, and then communicates between the tag and the anchor sequentially through time division, thereby ensuring accurate communication without delay or error between tags/anchors. It is possible to measure distance and thereby measure position and speed.

And, through this, the accuracy of predicting collisions and dangerous situations is improved. In addition, because the time of each moving object is synchronized, each moving object stores images for the same time, and it is easy to analyze the images for the same time for each moving object when a collision or dangerous situation occurs.

As described above, in one embodiment, after synchronizing the time between the tag of the smart gateway and the master gateway RTLS server, communication between the tag and the anchor is sequentially performed through time division.

That is, all tags are set to the same time by synchronizing the time of tags applied to multiple moving objects with the RTLS server of the master gateway, and each tag sequentially performs 1:1 communication with the anchor by time division.

Therefore, this eliminates communication delays and errors between tags/anchors and provides accurate distance measurement and location and speed measurement.

FIGS. 5A and 5B are diagrams for explaining in detail a time synchronization operation applied to an accident prevention system using deep learning according to an embodiment.

As shown in FIGS. 5A and 5B, time synchronization according to one embodiment first sets a communication method between a tag and an anchor. For example, the tag and the anchor perform 1:1 communication. And, if multiple tags attempt to communicate with the anchor, only one tag should communicate (see a).

Next, as shown in b, time synchronization is performed between the master gateway and the tag during initial system execution, and at this time, the master gateway transmits the time through broadcasting. And, distant tags are synchronized through the LoRa Mesh network.

Also, as in c, time synchronization for new moving objects is first classified based on Tag ID, and when a new moving object without time synchronization is detected, it is broadcasted to LoRa Mesh. Time synchronization for the entire system is achieved.

Additionally, d shows a mesh network configuration according to one embodiment.

Meanwhile, this system additionally has the following configuration. In other words, by performing time synchronization using the pull-push relationship below, time synchronization can be performed more quickly and efficiently.

To this end, this time synchronization configuration first performs time synchronization according to a pull-push relationship from the time synchronization format below in the master gateway, thereby synchronizing time in response to multiple different master/slave types (slave: relay gateway and Smart Gate Gay).

And, in this case, the time synchronization format is provided as follows.

a) First, the master gateway transmits a pulling signal to each slave in the form of a command and instructs the slave to report.

b) Then, the slave also has a user memory that records the pull-push relationship, and when sending out the pulling signal, it reports to the master by returning the pushing signal that passed through the user memory in an event-driven format.

On the other hand, in addition, the operation of performing the time synchronization is to read the user memory address when receiving the pulling signal from the slave, each time reporting through the user memory, and perform different pushing operations for each pull-push relationship within the user memory. The operation of continuously writing signals is performed repeatedly.

Also, in this case, when reporting through the user memory, time synchronization is performed according to the master slave type by setting the distance measurement format command differently for a plurality of different master slave types.

On the other hand, additionally, when such a system provides collision risk information through an administrator terminal, etc., it matches the database to the administrator terminal in real time and secures a connection, allowing the actual information to be delivered to the administrator quickly and easily.

To this end, the master gateway performs the following operations.

A) First, in order to match the database, when providing collision risk information, etc., a table storing the device registration information and data of the administrator terminal is provided identically to the administrator's (mobile) terminal, etc., and the table Register the matching relationship in advance.

B) When changing content in the table, it is synchronized to the other table according to the matching relationship.

C) When synchronizing the tables, the content is diversified for each data type for each type of notification information (or service and function), so that the databases match each other in real time.

A) Next, in this case, in order to secure the connection with the administrator terminal, first check whether the registered local communication network is connected, and if the local communication network is connected as a result of the confirmation, the corresponding Connect using the settings administrator public account.

B) As a result of the above confirmation, if the local communication network is not connected, it is secondarily checked whether the registered wireless communication network is connected.

C) As a result of the above confirmation, if the wireless communication network is connected, the connection is made with an individual IP address. On the other hand, when the wireless communication network is not connected, the connection is made using the terminal identification number of the registered mobile communication network, thereby securing a real-time connection with the administrator terminal.

On the other hand, when making a real-time connection with the administrator terminal, an IP table is used to monitor registered IPs and manage monitoring (or logs) of access by unauthorized persons to ensure security of the connection.

A) Specifically, for this purpose, an IP table is first configured in advance, registering the administrator public account of the local communication network and the individual IP (Internet Protocol) address of the wireless communication network.

B) Then, when an alarm is provided to the administrator terminal in this way, a HELLO message of the corresponding communication network is sent and the next hop switch IP address in the response result is extracted.

C) Next, check the switch IP address that is the same as this next hop switch IP address in the switch neighbor connection relationship list.

D) As a result of the above confirmation, if there is a switch IP address that is the same as the next hop switch IP address, it is checked whether the corresponding administrator public account or individual IP address is also in the IP table, and thus whether an unauthorized person is connected is checked.

E) As a result of the above confirmation, if the corresponding administrator public account or individual IP address is also present in the IP table, a JOIN/PRUNE message is sent, thereby connecting to the corresponding communication network.

On the other hand, when such a system provides collision risk information, it learns the information from the configuration below and can smoothly provide an alarm in advance.

In other words, in addition, this system provides good service by creating a learning model for monitoring by taking into account surrounding conditions or situations such as time synchronization that slightly differs for multiple different field locations and gateway device types.

In this case, these learning models attribute data to various field locations and time zones, thereby increasing the processing rate.

a) First, for this purpose, for example, when providing information about alarms, information including actual surrounding conditions or situations, such as time synchronization that varies slightly for multiple different field locations and gateway device types, is provided at the time and location. Define the learning model by classifying it as follows.

b) Next, extract basic datasets for a number of different surrounding states or situation information.

c) These datasets are then characterized to reflect multiple different locations, time periods, etc.

d) So, determine the properties of the state information for multiple different learning models based on these attribution results.

e) Then normalize the determined results.

f) Then, based on these normalization results, state information is set for multiple different learning models. Therefore, a number of different ambient state/situation information are used to set independent (alarm information) and dependent (ambient state/situation information) variables to provide alarm information.

g) Next, the above setting results are generated as learning and training data.

h) So, through this, a deep learning-based monitoring learning model is created from these results.

Therefore, when this system provides collision risk information as described above, it provides an alarm based on this content, thereby providing a service that actually helps.

1000: Accident prevention system using deep learning
100: Edge terminal (mobile device) 110: UWB tag module
120: LoRa T module 130: WiFi module
140: shooting module 150: display module
160: Alarm module 170: Information processing module
200: slave gateway 210: UWB anchor module
220: LoRa T module 230: Control module
300: Master gateway 310: WiFi module
320: LoRa G module 330: Ethernet module
340: Control module 400: Control server

Claims (4)

Tag-equipped moving objects (edge terminals), including construction machinery passing through the site;
Smart gateways installed in a plurality of different locations on the site to collect tag IDs and distance data obtained from tags for a plurality of different moving objects (edge terminals) for each moving object;
Relay gateways installed in a plurality of different sectors within the site according to the same data transmission range to relay tag ID and distance data for each corresponding moving object; and
a master gateway that receives tag IDs and distance data for each moving object in batches from the relay gateway, calculates the position and speed of each moving object, and provides information on the risk of collision with surrounding moving objects; It includes,

The master gateway is,
a) It is composed of the smart gateway and a short-range communication mesh, and the relay gateway is composed of a Wi-Fi mesh,
b) Perform time synchronization with each gateway by broadcasting the internal RTLS server time through the short-range communication mesh and the Wi-Fi mesh, and also perform time synchronization whenever a new moving object is detected while performing time synchronization. By repeatedly performing each broadcast again,
c) When the time synchronization is performed, the signal from the tag for each moving object is transmitted as distance data for each tag ID through the anchor to the Wi-Fi mesh,
d) analyzing the distance data for each tag ID in an RTLS server to calculate the position and speed for each moving object and providing information on the risk of collision with surrounding moving objects; Accident prevention system using deep learning characterized by . In claim 1,
The master gateway is
In the case of performing the time synchronization (b)), time synchronization is performed according to the pull-push relationship from the time synchronization format below, and time synchronization is performed in response to multiple different master/slave types (slave: relay gateway and Smart Gate Gay),
The time synchronization format is,
a) Sending a pulling signal in the form of a command for each multiple different slaves to instruct the slaves to report,
b) the slave has a user memory that records the pull-push relationship, and when transmitting the pulling signal, the slave returns a push signal that has passed through the user memory in an event-driven format and reports it to the master; Accident prevention system using deep learning characterized by . In claim 2,
The master gateway is
Whenever reporting through the user memory, reading a user memory address when receiving the pulling signal from the slave and repeatedly performing an operation of continuously writing different pushing signals for each pull-push relationship in the user memory; Accident prevention system using deep learning characterized by .







delete

Images