KR100973586B1 - Monitoring system in railway station for determining dangerous situation using neural network and method thereof - Google Patents

Abstract

A platform monitoring system and method are disclosed. According to the platform monitoring method of the present invention, a plurality of input patterns are input to a neural network learning pattern recognition for a dangerous situation in a platform, and the error is reduced by comparing the output values of the neural network for the input pattern with a predetermined target value. Neural network learning step of adjusting the weight of the neural network in the direction; And determining a risk situation in the platform using the learned neural network. According to the present invention, it is possible to make an intelligent and quick judgment on the risk situation in the platform and to increase the reliability of the judgment.

Train, platform, monitoring, neural network

Description

Platform monitoring system and method for determining dangerous situation in platform using neural network {MONITORING SYSTEM IN RAILWAY STATION FOR DETERMINING DANGEROUS SITUATION USING NEURAL NETWORK AND METHOD THEREOF}

The present invention relates to a platform monitoring system and a method thereof, and more particularly, to a platform monitoring system and method for determining a dangerous situation in the platform using a neural network.

Passenger safety is the most important task in the railroad sector. Even though railroad safety is centrally managed, dozens of passengers are being killed every year due to platform crash accidents, which is emerging as the most urgent problem to be solved in the railroad sector.

By the way, the conventional platform monitoring system, the worker must always monitor the situation (showing several divided screens sequentially according to the number of CCTVs) by simply installing an analog-type closed-circuit TV in a place to monitor. Therefore, while manpower is wasted, there is a problem in that it is difficult to quickly grasp and cope with a dangerous situation in a platform such as a passenger track crash, a door jamming, or a fire accident.

Therefore, it is possible to intelligently and quickly determine the types and the risks of the risk situations occurring in the platform, instead of the platform monitoring system that simply transmits the image information to the engineer or the station office in order to quickly identify and cope with the dangerous situation in the platform. Active platform monitoring system is required.

SUMMARY OF THE INVENTION An object of the present invention is to provide a platform monitoring system and method capable of making intelligent and rapid determination of a dangerous situation in a platform and increasing the reliability of the decision.

The object is, according to the present invention, input a plurality of input patterns to the neural network learning pattern recognition for the dangerous situation in the platform, and compares the output value of the neural network for the input pattern with a predetermined target value Neural network learning step of adjusting the weight of the neural network in the direction of error reduction; And determining a danger situation in a landing area using the learned neural network.

Here, the plurality of input patterns may be selected based on a result of analyzing the platform monitoring information expected according to the pattern of the dangerous situation.

The expected platform monitoring information may include at least one of thermal camera information, stereo camera information, sensor information, and train entry and exit information.

The neural network learning step may include: (a) initializing a weight of the neural network to an arbitrary value; (b) repeating steps (b1) to (b3) for each of the plurality of input patterns; (b1) checking an output value of the neural network with respect to the input pattern; (b2) calculating an error between the identified output value and a predetermined target value; And (b3) adjusting the weight based on the calculated error.

Step (b) may be repeated for all of the plurality of input patterns until the error is included in a preset tolerance range.

The risk situation determining step may include collecting platform monitoring information for a preset platform monitoring area; Inputting the collected platform monitoring information into the learned neural network; And determining a dangerous situation in the landing area based on the output result of the neural network.

In the collecting of the platform monitoring information, at least one of thermal camera information, stereo camera information, sensor information, and train entry and exit information may be collected.

The dangerous situation in the platform may include at least one of an object track crash, pinch of a passenger door, a fire, and a passenger safety line violation.

The object is, according to the present invention, input a plurality of input patterns to the neural network learning pattern recognition for the dangerous situation in the platform, and compares the output value of the neural network for the input pattern with a predetermined target value A neural network learner for adjusting a weight of the neural network in a direction in which an error is reduced; An information collector configured to collect platform monitoring information for a preset platform monitoring area; And an information processing unit for inputting the collected platform monitoring information to the learned neural network, and determining a risk situation in the platform based on the output result of the neural network.

Here, the plurality of input patterns may be selected based on a result of analyzing the platform monitoring information expected according to the pattern of the dangerous situation.

The expected platform monitoring information may include at least one of thermal camera information, stereo camera information, sensor information, and train entry and exit information.

The neural network learning unit (a) initializes the weight of the neural network to an arbitrary value, (b) for each of the plurality of input patterns (b1) confirms the output value of the neural network for the input pattern, ( b2) calculate an error between the identified output value and a predetermined target value, and (b3) adjust the weight based on the calculated error.

The neural network learner may adjust the weight until the error is included in a preset tolerance range for all of the plurality of input patterns.

The information collecting unit may include at least one of a thermal camera module, a stereo camera module, a sensor module, and a train detection module.

The dangerous situation in the platform may include at least one of an object track crash, pinch of a passenger door, a fire, and a passenger safety line violation.

The present invention, by using the neural network learned pattern recognition for the risk situation in the platform to determine the risk situation in the platform, it is possible to intelligently and quickly determine the risk situation in the platform, it is possible to increase the reliability of the determination .

In order to fully understand the present invention, the operational advantages of the present invention, and the objects achieved by the practice of the present invention, reference should be made to the accompanying drawings which illustrate preferred embodiments of the present invention and the contents described in the accompanying drawings.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the following description, well-known functions or constructions are not described in order to avoid unnecessary obscuration of the present invention.

1 is a schematic diagram of a platform monitoring system according to an embodiment of the present invention, Figure 2 is a conceptual diagram of the neural network of Figure 1, Figure 3 is a diagram for explaining the neural network learning unit of FIG.

Referring to FIG. 1, the boarding point monitoring system according to the present embodiment includes an information collecting unit 110 for collecting boarding point monitoring information for a preset boarding area monitoring area, and the boarding point monitoring information collected by the information collecting unit 110. Is applied to the neural network 130, the learning is completed, the information processing unit 120 to determine the dangerous situation in the platform, and the neural network learning unit 140 for learning the pattern recognition for the dangerous situation in the platform for the neural network 130 It is provided.

Here, the platform monitoring area may be set to a train area into which a train including a train track enters, an area around a safety line provided for the safety of passengers, and the like. However, in the present invention, the platform monitoring area is not limited to one specific area, but may be individually set according to the type and content of the platform monitoring information to be collected.

As shown in FIG. 1, the information collecting unit 110 collects platform monitoring information about an object that has fallen on a train track, a passenger caught in a door of a train, a fire in a platform, a passenger who has invaded a safety line, and the like. The thermal camera module 111, the stereo camera module 112, the sensor module 113 and the train detection module 114 is provided. That is, the platform monitoring information includes thermal camera information, stereo camera information, sensor information, and train entry and exit information. However, in the present invention, the information collecting unit 110 is not necessarily composed of the above four modules, of course, some of the four modules may be omitted or other modules may be added. On the other hand, if the module constituting the information collecting unit 110 is different, the type and content of the platform monitoring information is also different.

Thermal imaging camera module 111, it is preferable that two are installed in the front and rear of the landing so that the entire landing situation can be photographed. The thermal camera module 111 operates like a normal video camera module, but there is a difference in that the captured image detects infrared energy rather than general light. That is, the thermal camera module 111 is a type of infrared camera module is a camera module commonly used for the purpose of intruder monitoring and fire situation determination in the unmanned security system. The thermal camera module 111 detects infrared energy from an object to be photographed in the form of an analog wavelength and reproduces it as a temperature distribution image.

About 10 stereo camera modules 112 may be installed at predetermined intervals along the tracks of the train so as to photograph the train entering the platform and the entire platform situation. Unlike the general mono camera module, since the stereo camera module 112 can obtain three-dimensional coordinate data for each pixel of the acquired image, the platform monitoring area can be defined as an arbitrary three-dimensional space, and thus the platform monitoring area. It is possible to obtain three-dimensional coordinate data for an object in the body.

The sensor module 113 is responsible for monitoring the area around the train track, in particular, the area lower than the track height, and a plurality of sensor modules 113 are installed at predetermined intervals along the train track at the lower end of the platform to detect the entire train track. At this time, the sensor module 113 may be implemented as an infrared sensor, a laser sensor, an ultrasonic sensor, etc., but in this embodiment, the installation cost is low and the installation is easy, but the construction is implemented as an infrared sensor without problems in operating in the subway environment. do.

The train detection module 114 is a means for collecting information regarding the entry and / or departure of the train (hereinafter, referred to as 'train entry and exit information') in the platform. Implemented as a wireless network device capable of transmitting and receiving to each other. However, in the present invention, the train detection module 114 may be implemented in a general sensing method. Meanwhile, train entry and exit information may also be obtained through image difference analysis by the stereo camera module 112, but in this case, since the load of the entire system increases, train entry and exit information is collected as in the present embodiment. It is preferable that the train detection module 114 to be provided separately.

Referring to FIG. 1, the information processing unit 120 includes a neural network that completes learning of platform monitoring information such as thermal camera information, stereo camera information, sensor information, and train entry / exit information collected by the information collection unit 110 ( 130, and determines the dangerous situation in the landing based on the output result of the neural network 130. The information processing unit 120 is implemented in a server installed in the station or general control room of the corresponding platform.

In the present embodiment, the dangerous situation in the platform is a situation in which an object that impedes the operation of the train falls on the train track (hereinafter referred to as 'object track crash'), or a situation where a passenger is caught in the door gap (hereinafter referred to as 'passenger door gap'). This includes the situation of fires in the platform (hereinafter referred to as 'fire occurrence') and the situation where passengers have violated the safety line (hereinafter referred to as 'passenger safety line violation'). However, in the present invention, the dangerous situation in the landing can be changed or added without being limited to the above. Meanwhile, an object includes a passenger and an object.

Here, the determination of the risk situation in the platform means to determine the type of risk situation and the degree of danger, etc. occurring in the platform, where the risk refers to the severity of the risk situation and the urgency of countermeasures. It can be expressed as a level value in consideration.

In general, neural network 130 is composed of an input layer, a hidden layer and an output layer, each layer comprising a plurality of neurons. At this time, the neurons of the input layer and the neurons of the hidden layer are interconnected, the neurons of the hidden layer and the neurons of the output layer are interconnected, so that the output of each layer is transferred to the input of the next layer. On the other hand, the hidden layer may be composed of one or a plurality of layers.

In the present embodiment, the neural network 130 includes an input layer consisting of four neurons, one hidden layer consisting of five neurons, and an output layer consisting of four neurons, as shown in FIG. 2. do. Here, the input layer is inputted to the platform monitoring information such as thermal camera information, stereo camera information, sensor information, train entry and exit information, the output layer is in the platform such as object track crash, passenger door jamming, fire occurrence, passenger safety line violation It outputs a dangerous situation, and the hidden layer is located between the input layer and the output layer and receives the output of the input layer and passes it to the input of the output layer. However, in the present invention, the number of neurons constituting each layer and the number of hidden layers may be appropriately changed according to the environment in which the platform monitoring system is used, the required reliability, the characteristics of the input information, and the like.

On the other hand, the layers of the neural network 130 are correlated with each other according to the weight, and a predetermined correlation is formed between the input of the input layer and the output of the output layer by the weight. That is, the correlation between the input of the input layer and the output of the output layer is adjusted by modifying the weight. The weight is initially set to an arbitrary value, and is modified in a direction in which an error between the output value of the neural network 130 and the preset target value decreases through an iterative learning process. That is, the "learning process of the neural network 130" refers to a process of finding weights optimized for the system to which the neural network 130 is applied. On the other hand, the final learning result of the neural network 130 is mounted in the form of memory on the server on which the information processing unit 120 is implemented. Meanwhile, the weight includes a first weight that determines a correlation between the hidden layer and the output layer, and a second weight that determines a correlation between the input layer and the hidden layer.

The platform monitoring system according to the present embodiment is operated in the actual platform in a state where the neural network 130 is trained by the neural network learning unit 140.

1 and 3, the neural network learning unit 140 is a means for learning or constructing a neural network circuit under optimum conditions before actually operating the platform monitoring system according to the present embodiment. When the learning process of the neural network 130 is performed in the form of a field test, the neural network learning unit 140 is implemented in a server installed in the station control room or the general control room of the corresponding platform as in the above-described information processing unit 120. If the learning process of the neural network 130 is performed in the form of a simulation test, it is implemented in a server installed in the laboratory.

The neural network learning unit 140 inputs a plurality of input patterns to the neural network 130 that learns pattern recognition for a dangerous situation in a platform, and sets a predetermined target value of an output value of the neural network 130 for each input pattern. The neural network 130 is trained by adjusting the weight of the neural network 130 in a direction in which the error is reduced in comparison with.

Here, the plurality of input patterns input to the neural network 130 are selected based on the results of analyzing the platform monitoring information such as thermal camera information, stereo camera information, sensor information, and train entry and exit information expected according to a dangerous situation. do. At this time, the platform monitoring information expected according to the risk situation can be obtained through the classification of the data obtained from the simulation test or field test for the platform monitoring system.

Table 1 below shows an example of a plurality of input patterns used for learning of the neural network 130 in the platform monitoring system according to the present embodiment.

TABLE 1

Thermal Camera Information Stereo Camera Information Sensor Information Train Entry and Exit Information Behavior One One One One Passenger track crash One One One 0 Passenger track crash 3 0 0 0 Fire 0 One One 0 Object line crash 0 One One 0 Crashed 2 2 0 One Passenger door jamming 2 2 0 0 Passenger safety line violation ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...

As illustrated in Table 1, the plurality of input patterns are allocated to the neural network 130 by being assigned an integer equal to or greater than '0' for each platform monitoring information. Here, the meaning of the integer assigned to each platform monitoring information is shown in Table 2 below. Meanwhile, in Table 1, the target value is expressed as a kind of risk situation, but is actually set to an assigned value according to the kind of risk situation and its risk.

TABLE 2

Thermal Camera Information Stereo Camera Information Sensor Information Train Entry and Exit Information 0 No occurrence No occurrence No occurrence No occurrence One Track crash Track crash Occur Train entry 2 Safety line violation Door jamming ... Train 3 Fire ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...

When a plurality of input patterns as exemplified in [Table 1] and [Table 2] are selected, the neural network unit, as shown in FIG. 3, first initializes the weight of the neural network 130 to an arbitrary value. do. Next, the neural network learning unit 140 checks an output value of the neural network 130 for each input pattern with respect to each of the plurality of input patterns, calculates an error between the identified output value and a predetermined target value, and calculates the calculated error. Adjust the weight based on Here, the neural network learner 140 repeats the learning process of adjusting the weight until the error is included in a preset tolerance range for all the plurality of input patterns. At this time, if the error for all the plurality of input patterns are included in the tolerance range, the neural network 130 is recognized that the learning is completed. On the other hand, the learning completion condition of the neural network 130 may be changed according to the required reliability, etc. Unlike the present embodiment based on the tolerance range, for example, the sum of the errors calculated for each input pattern is a reference value. It may be configured to repeat the learning process until:

As described above, the platform monitoring system according to the present embodiment analyzes platform monitoring information such as thermal camera information, stereo camera information, sensor information, and train entry and exit information expected according to a dangerous situation, and based on the neural network 130 Learning training for a certain period of time, it is possible to build a neural network 130 having a weight optimized for determining a risk situation in the platform.

Meanwhile, a method for calculating an error and a method for correcting a weight in the learning process of the neural network 130 will be described in parallel with the platform monitoring method by the platform monitoring system according to the present embodiment described below. .

4 is a flowchart illustrating a platform monitoring method by the platform monitoring system of FIG. 1, and FIG. 5 is a flowchart illustrating a neural network 130 learning step of FIG. 4.

Referring to FIG. 4, the platform monitoring method according to the present exemplary embodiment may include learning a pattern recognition for a dangerous situation in a platform with respect to the neural network 130 (S100), and using the learned neural network 130. Determining a dangerous situation in the platform (S200).

The neural network 130 learning step S100 will be described in detail with reference to FIG. 5 as follows.

First, a step of selecting a plurality of input patterns based on the result of analyzing the expected platform monitoring information according to the pattern of the dangerous situation in the platform (S110). In the present embodiment, the platform monitoring information includes thermal camera information, stereo camera information, sensor information, and train entry and exit information for a predetermined platform monitoring area.

Next, the step of initializing the weight of the neural network 130 to an arbitrary value is performed (S120). For example, the weight of neural network 130 is initialized to any value of -1 and 1 phase.

Next, for each of the selected plurality of input patterns, the output value of the neural network 130 for the input pattern is checked (S131), and the error is calculated by comparing the identified output value with a preset target value (S132). In operation S130, a weight is adjusted based on the calculated error.

Here, the error for the neuron k of the output layer is calculated by Equation 1 below, and the first weight for determining the correlation between the hidden layer and the output layer is corrected by Equation 2 below. .

[Equation 1]

E _k = (t _k -o _k ) × o _k (1-o _k )

[Equation 2]

w _jk + = L × E _k × o _j

E _k : error for neuron k in the output layer

o _k : Output of neuron k in the output layer

t _k : target value

w _jk : first weight between neuron j of hidden layer and neuron k of output layer

L: A small positive number representing the degree of learning

o _j : Output of neuron j in hidden layer

In addition, the error for the neuron j of the hidden layer is calculated by Equation 3 below, and the second weight for determining the correlation between the input layer and the hidden layer is corrected by Equation 4 below. .

&Quot; (3) "

E_k w_jk E _j = o _j (1-o _j )

E _k w _jk

&Quot; (4) "

w _ij + = L × E _j × o _i

E _j : error for neuron j in the hidden layer

o _j : Output of neuron j in hidden layer

n: number of neurons in the output layer

w _ij : second weight between neuron i of input layer and neuron j of hidden layer

L: A small positive number representing the degree of learning

o _i : Output of neuron i in input layer

Next, if it is determined that the above-described step S130 is performed for all input patterns, a step of determining whether an error calculated for all input patterns is included in a preset tolerance range is performed (S140). As a result of the determination, when it is out of the tolerance range, step S130 is performed again on the selected plurality of input patterns, and when it is included in the tolerance range, it is defined that the learning of the neural network 130 is completed. That is, step S130 of adjusting the weight of the neural network 130 is repeated until the errors for all input patterns are included in the tolerance range.

When the neural network 130 learning step (S100) as described above is completed, as illustrated in FIG. 4, a step of determining a dangerous situation in the landing using the neural network 130 that has been learned is performed (S200).

The risk situation determination step (S200), the step of collecting platform monitoring information such as thermal camera information, stereo camera information, sensor information, train entry and exit information (S210), and the completed network monitoring information learning neural network 130 In step (S220) and the step (S230) of determining the dangerous situation in the platform, such as object line fall, passenger door pinch, fire occurrence, passenger safety line violation, based on the output result of the neural network (130) It is completed by performing with.

As described above, the platform monitoring system and method according to the present embodiment, by using the neural network 130 learned pattern recognition for the risk situation in the platform by determining the risk situation in the platform, Intelligent and quick judgment can be made, and the reliability of the judgment can be improved.

It is apparent to those skilled in the art that the present invention is not limited to the above-described embodiments, and that various modifications and changes can be made without departing from the spirit and scope of the present invention. Therefore, such modifications or variations will have to be belong to the claims of the present invention.

1 is a schematic diagram of a platform monitoring system according to an embodiment of the present invention.

FIG. 2 is a conceptual diagram of the neural network of FIG. 1.

FIG. 3 is a diagram for describing a neural network learner of FIG. 1.

4 is a flowchart illustrating a boarding point monitoring method by the boarding point monitoring system of FIG. 1.

5 is a flowchart illustrating a neural network learning step of FIG. 4.

<Explanation of symbols for the main parts of the drawings>

110: information collector

120: information processing unit

130: neural network

140: neural network learning unit

Claims (15)

The neural network inputs a plurality of input patterns to a neural network learning pattern recognition for a dangerous situation in a platform, and compares an output value of the neural network with respect to the input pattern with a preset target value to reduce the error. Neural network learning step of adjusting the weight of the; And Determining a risk situation in a platform using the learned neural network; The plurality of input patterns are selected based on the result of analyzing the platform monitoring information expected according to the risk situation pattern, The expected platform monitoring information is characterized in that it includes thermal camera information, stereo camera information, sensor information and train entry and exit information, The neural network learning step may include: (a) initializing a weight of the neural network to an arbitrary value; (b) repeating steps (b1) to (b3) for each of the plurality of input patterns; (b1) checking an output value of the neural network with respect to the input pattern; (b2) calculating an error between the identified output value and a predetermined target value; And (b3) adjusting the weight based on the calculated error, The dangerous situation in the platform, the platform monitoring method, characterized in that it comprises at least one of the object track crash, passenger door pinch, fire occurrence, passenger safety line violation. delete delete delete The method of claim 1, In step (b), The platform monitoring method, characterized in that for all of the plurality of input patterns is repeated until the error is included in the preset tolerance range. The method of claim 1, The risk situation determination step, Collecting platform monitoring information for a preset platform monitoring region; Inputting the collected platform monitoring information into the learned neural network; And A platform monitoring method comprising the step of determining a dangerous situation in the platform based on the output of the neural network. The method of claim 6, Collecting the platform monitoring information, Platform monitoring method comprising collecting at least one of thermal imaging information, stereo camera information, sensor information and train entry and exit information. delete The neural network inputs a plurality of input patterns to a neural network learning pattern recognition for a dangerous situation in a platform, and compares an output value of the neural network with respect to the input pattern with a preset target value to reduce the error. Neural network learning unit for adjusting the weight of the; An information collector configured to collect platform monitoring information for a preset platform monitoring area; And An information processor configured to input the collected platform monitoring information into the learned neural network, and determine a risk situation in the platform based on an output result of the neural network; The plurality of input patterns are selected based on a result of analyzing the platform monitoring information expected according to the risk situation pattern, The expected platform monitoring information includes thermal camera information, stereo camera information, sensor information and train entry and exit information, The neural network learning unit (a) initializes the weight of the neural network to an arbitrary value, (b) for each of the plurality of input patterns (b1) confirms the output value of the neural network for the input pattern, ( b2) calculating an error between the identified output value and a predetermined target value, and (b3) adjusting the weight based on the calculated error, The dangerous situation in the platform, the platform monitoring system, characterized in that it comprises at least one of the object track fall, passenger door pinch, fire occurrence, passenger safety line violation. delete delete delete 10. The method of claim 9, The neural network learning unit, The platform monitoring system, characterized in that for all of the plurality of input patterns the weight is adjusted until the error is included in a predetermined tolerance range. 10. The method of claim 9, The information collecting unit, Platform monitoring system comprising at least one of a thermal imaging camera module, a stereo camera module, a sensor module and a train detection module. delete

Images