JPH07228250A - Intrack monitoring device and platform monitoring device - Google Patents

Abstract

PURPOSE:To perform decision of the content of an obstacle in a track by providing an alarm outputting means to output an alarm when a detected object is picked up, and a display means to display the assortment of the detected object when the assortment of the detected object is specified by a detected object specifying means. CONSTITUTION:Digital image data 1d regenerated by a memory means 53 and drop attention information are outputted to a monitor 6 by a display control means 55. The digital image data 1d, i.e., an image photographed by a camera 3 and drop attention information are displayed by the monitor 6. A station officer decides through alarming of an alarm buzzer 7 that the detected object is detected. The detected object is confirmed by the monitor 6 and the station officer urges persons on a platform 1 to pay attention and the erroneous drop of a person on the platform 1 onto a truck 2 is prevented from occurring.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention monitors an in-track or platform of a railroad to detect obstacles existing in the track at an early stage, and at the same time, monitors in-orbit to predict the fall of people and objects from the platform. Device and platform monitoring device.

[0002]

2. Description of the Related Art A method using a piezoelectric sensor is known as a method for detecting an obstacle that has fallen into a track. This technology uses obstacles that fall from the platform into the orbit,
The obstacle that has entered is placed on the piezoelectric sensor installed near the track and a load is applied, and it is determined that the obstacle exists in the track.

In the technique disclosed in Japanese Patent Laid-Open No. 63-235163, a trajectory is monitored by using a camera, and an obstacle in the trajectory is detected from an image captured by the camera by using an image processing technique. Is the way.

The above-mentioned method for detecting obstacles in a track can notify a station staff or a train crew by sounding an alarm buzzer or the like when an obstacle is detected, thereby preventing an accident from occurring in a train in advance. It is a thing.

[0005]

In the obstacle detection method using the piezoelectric sensor and the obstacle detection method using image processing described above, it was possible to notify the station staff or train crew when an obstacle was detected. However, since the contents of obstacles existing in the orbit cannot be determined, only similar detection can be performed, for example, when a small animal crosses the orbit and when a human falls from the platform. It was

With the recent increase in the number of passengers of electric vehicles, the number of passengers on the platform waiting for the electric vehicles also increases, and the possibility of accidentally falling from the platform into the track is increasing. There has been a demand for a technology for predicting and preventing the fall into the orbit from such a platform, but in the obstacle detection method using the piezoelectric sensor and the obstacle detection method using image processing described above, It only detects obstacles that have fallen into.

Therefore, an object of the present invention is to provide an in-orbit monitoring device and a platform monitoring device for judging the contents of obstacles in the orbit and predicting and preventing the obstacles falling from the platform.

[0008]

In order to achieve the above-mentioned object, the invention according to claim 1 inputs an image from a camera which has photographed the inside of the orbit, and obtains a difference image with a predetermined initial image, The detected object is extracted from this difference image, the distance from the position of the detected object on the difference image to the point where the camera and the detected object actually exist is estimated, and this distance and the number of pixels of the detected object on the difference image are estimated. , Comparing the actual size of the detected object, the actual size of the detected object estimated by this calculating means, and a predetermined first threshold value to determine the type of the detected object. When the detection object specifying means for specifying and the detection object is extracted by the computing means, an alarm output means for outputting an alarm, and when the type of the detection object is specified by the detection object specifying means, the type of the detection object is displayed. And a display means for displaying.

According to the second aspect of the present invention, when the number of pixels of the detection object extracted by the calculation means becomes equal to or larger than a second threshold value set in advance, it is determined that the train is approaching. Means and image input stop means for stopping the input of the image captured by the camera to the calculation means when the train approach determination means determines that the train is approaching.

According to the third aspect of the present invention, the camera for photographing the platform and the image photographed by the camera are input to obtain a difference image between a predetermined initial image and the detected object is extracted from the difference image. And a judgment means for judging whether or not the detection object extracted by this calculation means exists at the edge of the platform, and if this judgment means judges that the detection object exists at the edge of the platform, it goes under the platform. And a display means for inputting and displaying the prediction result of the first prediction means.

According to a fourth aspect of the invention, the distance between the camera and the point where the detected object actually exists is estimated from the position of the detected object on the difference image provided in the calculation means, and this distance Detected object specifying means for specifying the type of the detected object from the number of pixels of the detected object on the difference image, measuring means for measuring the number of detected objects of the same type specified by the detected object specifying means, and this measuring means The number of detected objects measured by
A second predicting means for predicting a fall under the platform when the preset number is exceeded and instructing the alarm output means to output an alarm and for inputting the prediction result to the display means may be provided. Good.

[0012]

With the above-mentioned structure, in the invention according to claim 1,
The orbit is photographed by the camera, and the calculation means extracts the detected object from the difference image between the initial image and the image photographed by the camera. Then, the distance from the position of the detected object on the difference image to the point where the camera and the detected object actually exist is estimated, and the actual size of the detected object is estimated from this distance and the number of pixels of the detected object. The detected object specifying means compares the actual size of the detected object with a predetermined first threshold value.

The first threshold is, for example, the width and height of a human being or the like. Therefore, the type of the detected object can be specified by comparing the actual size of the detected object with the first threshold value. The fact that the detection object has been extracted by the computing means means that there is an obstacle in the running of the train in the track, so an alarm is output from the alarm output means. Station staff can judge that there is an obstacle in the track by this warning. or,
Since the type of the detected object specified by the detected object specifying means is displayed on the display means, the station staff can know the obstacle existing in the track by this display and can take safety measures.

Further, in the invention according to claim 2, the train approach determination means compares the number of pixels of the detected object with a predetermined second threshold value, and when it becomes equal to or larger than the second threshold value. It is judged that the train is approaching. When the train approaches, the inside of the track is covered with the train, so the image input from the camera to the calculation means is stopped by the image input stopping means.

According to the third aspect of the present invention, the camera photographs the platform, and the computing means extracts the detected object from the difference image between the initial image and the image photographed by the camera. If it is judged by the judging means whether or not the position of the detected object is at the end of the platform, and if it is judged that the detected object exists at the end of the platform, the first predicting means may cause the detected object to fall into the orbit from the platform. Predict that there is.
An alarm is output from the alarm output means, and the prediction result is displayed on the display means, so that the station staff can take safety measures.

In the invention according to claim 4, the detected object specifying means estimates the actual size of the detected object and specifies the type of the detected object. For example, the number of detection objects identified as a human is measured by the measuring means. Then, when the number of detected objects exceeds a predetermined number, it is predicted by the second predicting means that the detected objects may fall into the track from the platform.
Then, an alarm is output from the alarm output means and the prediction result is displayed on the display means, so that the station staff can take safety measures.

[0017]

An embodiment of the present invention will be described in detail with reference to the drawings. 1 to 5 are views showing an embodiment of the present invention.
1 is an overall configuration diagram of an in-orbit monitoring device and a platform monitoring device, FIG. 2 is a flowchart showing a processing flow of the in-orbit monitoring device and the platform monitoring device, FIG. 3 is an explanatory diagram of labeling processing, and FIG. FIG. 5 is a diagram for explaining an actual size estimation method, and FIG. 5 is a diagram showing a processing flow of the image processing apparatus.

In FIG. 1, 1 is a platform (hereinafter referred to as a platform), 2 is a track, and 3 is a platform 1 and a track 2.
Multiple cameras to shoot. The camera switching device 4 selects one camera from the plurality of cameras 3, and the selected camera 3 outputs captured image data (hereinafter referred to as analog image data) I _a to the image processing device 5. A monitor device (hereinafter referred to as a monitor) 6 for displaying the image processing result and an alarm buzzer 7 are connected to the image processing device 5. The image processing device 5 includes a control unit 51, an analog-digital conversion unit (hereinafter referred to as A / D conversion unit) 52, a storage unit 53, a calculation unit 54, and a display control unit 55.

The control means 51 outputs various commands in the image processing calculation process, which will be described later, and controls the image processing apparatus 5 as a whole. The analog image data I _a output from the camera 3 is input to the A / D conversion means 52, and digital image data I _d (for example, black and white 16 gradations, black is represented by “0”, white is represented by “1”). Pixel value).

The storage means 53 stores and reproduces various image data based on a command from the control means 51.
And initial image data I _i of the trajectory 2 and mask image data M described later are stored in advance. The storage means 53 also stores the digital image data I _d converted by the A / D conversion means 52 and the processing data C in the image processing calculation process.

The calculation means 54 is for performing image processing calculation on various image data reproduced by the storage means 53 based on a command from the control means 51. Then, the control means 51 determines the result of the image processing calculation in the calculation means 54. Display control means 5
5 is a monitor device 6 based on the determination result of the control means 51.
It controls the display data to be displayed.

The flow of processing of the in-orbit monitoring device and the platform monitoring device thus configured will be described in detail with reference to the flowchart of FIG. The camera 3 selected by the camera switching device 4 photographs the home 1 and the trajectory 2, and the image processing device 5 inputs the analog image data I _a captured by the camera 3 (S1). The analog image data I _a is converted into digital image data I _d by the A / D conversion means 52 (S2). This digital image data I _d is control means 5
It is stored in the storage means 53 according to the write command from 1.

Then, in response to a read command from the control means 51, the storage means 53 reproduces the prestored initial image data I _i and digital image data I _d and outputs them to the computing means 54. The calculation means 54 uses the initial image data I _i.
And the digital image data I _d are subjected to difference processing to obtain difference image data C1 (S3). The differential image data C1 is stored in the storage unit 53 according to a write command from the control unit 51.

Next, in response to a read command from the control means 51, the storage means 53 stores mask image data M1 in the orbit that has been stored in advance (image data with "1" inside the orbit 2 and "0" outside the orbit 2). ) And the differential image data C1 are reproduced and output to the calculation means 54. The calculation means 54 performs a calculation for multiplying the difference image data C1 by the mask image data M1 in the track, and calculates the total value of each pixel value in the track 2. The control means 51 compares the predetermined value with the total value of the pixel values and determines that the train has approached the vicinity of the platform 1 when the total value of the pixel values becomes equal to or larger than the predetermined value (S4).

When the control means 51 determines that the train is approaching, the control means 51 directs the camera 3 to the image processing apparatus 5.
To output the analog image data I _a to the image processing device 5 (S).
5).

When the control means 51 determines that the train is not approaching, the control means 51 outputs a read command,
The storage unit 53 reproduces the stored difference image data C1 and outputs it to the calculation unit 54. The arithmetic means 54 performs a binarization process, compares it with a predetermined threshold value, and if it is equal to or larger than the threshold value, it is "1" (white), and if it is less than the threshold value, it is "0" (black). Is obtained (S6). Binary image data C
2 is a storage means 5 in response to a write command from the control means 51.
3 is stored.

Next, in response to a read command from the control means 51, the storage means 53 reproduces the stored binary image data C2 and outputs it to the calculation means 54. The calculation means 54 is “1”
A cluster in which (white) is continuously arranged is detected, and a labeling process is performed in which each detected cluster (hereinafter referred to as a detected object) is sequentially numbered to perform separation for each detected object (S7). For example, scanning the binary image data C2 from the upper left,
When (white) is detected first, it is numbered as "I", and when "1" is detected continuously in the horizontal direction, it is also numbered as "I". Number "I" and then number "II" when "1" (discontinuous on the side) is detected. When this is repeated and the scanning of one column is completed, the process returns to the left end of the second column and the same processing is performed. Specifically, the result of performing the labeling process by scanning the first column of the binary image data C2 as shown in FIG. 3A is as shown in FIG. 3B. When scanning from left to right, "1" is first detected in "C1", so "I" is added to "C1". Next, "G1" detects "1",
Since it is not continuous with "C1", add "II" to "G1". In the second and subsequent columns, "1" that continuously exists vertically and horizontally is detected. In this case, the problem is how to handle the diagonally continuous "1", but the diagonally continuous "1" is initially assigned a new number "III" depending on the pixel value to be scanned next. Change the number. Specifically, FIG. 3 (c)
In "B2", it is detected as "1", but since "1" of "B2" and "I" of "C1" are only diagonal continuation,
Add "III" to "B2". Then, "1" is detected by "C2". Since "1" of "C2" and "I" of "C1" are vertically continuous, "I" is added to "C2". "C
"I" of "2" and "III" of "B2" are continuous horizontally, so "I" is reattached to "B2". This process is repeated. Finally, the binary image data C2 in FIG.
The labeling process is performed as in (d), and the detection objects are separated.

If there is a numbered area in this labeling process, the control means 51 determines that there is a detected object (S8). When the control means 51 determines that there is no detected object,
The calculation means 54 ends the calculation. When the control means 51 determines that there is a detected object, the detected object is specified (S9).

Identification of the detected object will be described below. The distance from the camera 3 to the actual detected object is estimated from the position of the detected object on the labeling data C3, and the distance from the camera 3 to the detected object and the number of pixels of the detected object on the labeling data C3 are used to detect the detected object. Estimate the actual height and width. Then, the detected object is specified by comparing the actual height and width of the detected object with a predetermined threshold value. As shown in FIG.
If the position of the lowest pixel of the detected object I is y, the distance d from the camera 3 to the detected object can be estimated as d = 1 / (k ₁ × y × k ₂ ) (1). However, k ₁ and k ₂ are constants depending on the mounting height and angle of the camera 3.

From the distance d obtained from the equation (1), lev = 1 / [k ₃ / (d + k ₄ )] (2) is used to estimate the ratio lev of the actual size of one pixel on the labeling data C3. . However, k ₃ and k ₄ are camera 3
It is a unique constant. Further, the ratio lev obtained from the equation (2)
From H = lev × h W = lev × W (3), the actual height H and width W of the detected object are estimated. However, h is the height of the detected object on the labeling data C3, and w is the width of the detected object on the labeling data C3. By comparing the actual height H and width W of the detected object obtained by the equation (3) with predetermined threshold values, the control means 51 is obtained.
Can specify the detected object. However, the threshold is human, luggage,
It depends on other things.

Hereinafter, the case where the threshold value is set for a human and a human is specified among the detected objects will be described. Of the detected objects on the labeling data C3, those other than those identified as humans are removed by the computing means 54 to obtain the labeling data C4. This labeling data C4 is the control means 51.
It is stored in the storage means 53 according to a write command from.
Then, in response to a read command from the control means 51, the storage means 53 stores mask image data M2 in the home 1 (image data in which the inside of the home 1 is "1" and the outside of the home 1 is "0") stored in advance. The labeling data C4 is reproduced and output to the calculating means 54. The calculation means 54 uses the labeling data C4 as mask image data M in the home 1.
Multiply by 2. Then, the control means 51 determines whether or not there is a person in the home 1 (S10).

When the control means 51 determines that there is a person in the home 1 (S10), the control means 51 outputs a writing means command, and the storage means 53 outputs the labeling data C4.
Then, the on-home data C5 obtained by multiplying the mask image data M2 in the home 1 is stored. Further, the control means 51 outputs a read command, and the storage means 53 stores mask image data M3 of the edge portion of the home 1 stored in advance (an image in which the edge portion of the home 1 is "1" and other portions are "0"). (Data) and the on-home data C5 are reproduced and output to the calculation means 54. The calculation means 54 multiplies the on-home data C5 by the mask image data M3 of the end portion of the home 1. Then, the control means 51 determines whether or not there is a person at the end of the home 1 (S1).
1).

When the control means 51 determines that there is a person at the end of the home 1 (S11), the control means 51 outputs an alarm command to the alarm buzzer 7 and sounds the alarm buzzer 7 (S14). . Further, the control means 51 reproduces the information that a person is present at the end of the home 1 and a read command, and the storage means 53 reproduces the stored digital image data I _d and outputs it to the display control means 55.

The display control means 55 receives information from the control means 51 that there is a person at the end of the home 1, and displays fall caution information such as graphic data for displaying "fall caution" on the monitor 6, for example. create. Then, the display control means 55 outputs the digital image data I _d reproduced by the storage means 53 and the drop caution information to the monitor 6.
The monitor 6 displays the digital image data I _d , that is, the camera 3
The image taken in step S3 and the drop caution information are displayed (S1).
5). The station staff can judge that there is an object detected by the alarm buzzer 7 sounding, check the monitor 6 to call attention to the people on the platform 1, and prevent people on the platform 1 from accidentally falling on the track 2. .

If the control means 51 does not determine that there is a person at the end of the home 1 (S11), the control means 51 outputs a read command and the storage means 53 reproduces the stored on-home data C5. And outputs it to the calculation means 54. The calculation means 54 calculates the number of detected objects numbered by the labeling process, here the number of detected people. The control means 51 compares a predetermined number with the number of detected persons (S12), and if the number of detected persons is equal to or more than the predetermined number, the control means 51 outputs an alarm command to the alarm buzzer 7,
The alarm buzzer 7 is called (S14). Further, the control means 51 reproduces the stored digital image data I _d and outputs it to the display control means 55, with information and a read command that there are more than a predetermined number of people on the home 1.
The display control means 55 receives information from the control means 51 that there are more than a predetermined number of people on the home 1, and displays fall caution information such as graphic data for displaying "fall caution" on the monitor 6, for example. create.

The display control means 55 outputs the digital image data I _d reproduced by the storage means 53 and the drop caution information to the monitor 6. The monitor 6 displays the digital image data I _d , that is, the image taken by the camera 3 and the drop caution information (S15). The station staff can judge that there is an object detected by the alarm buzzer 7 sounding, check the monitor 6 to call attention to the people on the platform 1, and prevent people on the platform 1 from accidentally falling on the track 2. .

When the control means 51 determines that the number of detected persons is less than the specified value (S12), the control means 51 outputs a read command, and the storage means 53 reproduces the stored digital image data I _d to display control means. Output to 55. The display control means 55 monitors the digital image data I _d for the monitor 6
Output to. The monitor 6 displays the digital image data I
_d , that is, the image taken by the camera 3 is displayed (S1
5).

When the control means 51 does not determine that there is a person on the home 1 (S10), the control means 51 outputs a read command and the storage means 53 stores the mask image in the trajectory 2 stored in advance. The data M1 and the labeling data C4 are reproduced and output to the calculating means 54. The calculation means 54 multiplies the labeling data C4 by the mask image data M1 in the trajectory 2. Then, the control means 51 determines whether or not there is a person in the trajectory 2 (S13). Control means 51
When it is determined that there is a person in the trajectory 2, the control means 5
1 outputs an alarm command to the alarm buzzer 7 and sounds the alarm buzzer 7 (S14).

The control means 51 reproduces the digital image data I _d stored in the storage means 53 and outputs it to the display control means 55, together with the information that a person is in the trajectory 2 and the read command. The display control means 55 receives the information from the control means 51 that a person is present in the trajectory 2 and determines that the falling object is a human, and displays, for example, "a person falls in the trajectory / attention" on the monitor 6. The contents of falling objects and the existence information of falling objects such as graphic data are created.

Then, the display control means 55 outputs the digital image data I _d reproduced by the storage means 53 and the content / falling object existence information to the monitor 6. The monitor 6 displays the digital image data I _d , that is, the image photographed by the camera 3, and the content / falling object existence information of the falling object (S).
15). The station staff can judge that there is an object detected by the alarm buzzer 7 sounding, and can confirm it on the monitor 6 and take safety measures such as outputting a train stop signal.

Next, from the image input by the camera 3, the home 1
And the process of extracting the detected substances existing in the orbit 2 is shown in FIG.
It demonstrates using (a) -FIG.5 (g). The image data stored in advance in the storage unit 53 is shown in FIGS.
It is (d). However, although actually stored in the storage means 53 as digital image data, FIG. 5 shows an analog image.

FIG. 5A shows the initial image data I _i , FIG.
(B) is the mask image data M1 in the trajectory 2, FIG. 5 (c)
Is the mask image data M2 in the home 1, and FIG. 5D is the mask image data M3 at the end of the home 1. The image data obtained in the following process is actually digital image data, but FIG. 5 shows an analog image. When the camera 3 captures the analog image data I _a shown in FIG. 5E, the calculation means 54 performs difference processing,
The difference image data C1 shown in (f) is obtained.

The difference image data C1 is multiplied by the mask image data M1 in the track 2 shown in FIG. 5B in order to determine the approach of the train. Therefore, when the train is approaching, the train exists in the image data obtained by multiplying the difference image data C1 by the mask image data M1 in the track 2.

Therefore, the control means 51 compares the total value of the respective pixel values in the track 2 with a predetermined value, and when it is equal to or larger than the predetermined value, it is determined that the train is approaching. Difference image data C
When it is determined that the train does not exist in the image data obtained by multiplying 1 by the mask image data M1 in the track 2 and the train does not approach, the difference image data C
The binarization processing of 1 is performed to obtain the binary image data C2 shown in FIG. Further, in the calculation means 54, 2 shown in FIG.
The labeling process is performed on the value image data C2 to obtain the labeling data C3, and the actual height H and width W of the detected object are estimated using the equations (1) to (3).

Then, the actual height H, width W and the threshold value are compared by the control means 51 to specify the detected object. Then, for example, when the threshold value is set for a human, the labeling data C4 is obtained by removing the detected objects other than the human by the calculation means 54. This labeling data C4
, The mask image data M in the home 1 shown in FIG.
2 is multiplied by the calculation means 54 to obtain the on-home data C5, and the detected substance existing on the home 1 can be extracted. Further, the calculation means 54 multiplies the on-home data C5 by the mask image data M3 of the end portion of the home 1 shown in FIG. 5D, and when the detected object exists at the end portion of the home 1, the detected object is extracted. The control means 51 determines that there is a detected object at the end of the home 1.

In the home data C5, the home 1
It is possible to predict the number of the detected objects existing in the home 1 on the track 2 by calculating the number of the detected objects existing in the calculation unit 54 and comparing the calculated number with the specified value in the control unit 51. Further, the labeling data C4 can be multiplied by the mask image data M1 in the trajectory 2 shown in FIG.

In this way, the detected objects on the home 1 and the detected objects on the track 2 are extracted, the alarm buzzer 7 is used to notify the station staff, and the station staff displays the image taken by the camera 3 and the monitor 6 displaying the caution information. By checking, you can take safety measures.

Note that the automatic guide device may interlock with the monitor 6 to call attention. Further, the curved home 1 can also be combined with a plurality of cameras, and the camera 3 can be attached to the home 1 side and the wall side to prevent the shooting area from leaking.

[0049]

As described above, according to the first to fourth aspects of the present invention, since it is possible to monitor the home and the inside of the orbit by the camera and identify the contents of the detected object, In particular, by determining the position and density of people, it is possible to prevent them from falling into the orbit, and if a person, etc., falls into the orbit from the platform, it can be detected early and safety measures can be taken. It is possible to provide an in-orbit monitoring device and a platform monitoring device capable of taking advantage of the above.

[Brief description of drawings]

FIG. 1 is an overall configuration diagram of an in-orbit monitoring device and a platform monitoring device showing an embodiment of the present invention.

FIG. 2 is a flowchart showing a processing flow of the in-orbit monitoring device and the platform monitoring device shown in FIG.

FIG. 3 is an explanatory diagram of labeling processing.

FIG. 4 is a diagram for explaining a method of estimating the actual size of a detected object.

FIG. 5 is a diagram showing a flow of processing of the image processing apparatus.

[Explanation of symbols]

1 ... Home, 2 ... Orbit, 3 ... Camera, 4 ... Camera switching device, 5 ... Image processing device, 6 ... Monitor, 7 ... Alarm buzzer,
51 ... Control means, 52 ... A / D conversion means, 53 ... Storage means, 54 ... Calculation means, 55 ... Display control means.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Satoshi Kojima 1-1-1, Shibaura, Minato-ku, Tokyo Inside Toshiba Headquarters office (72) Kenjiro Naemura 1st Toshiba Town, Fuchu City, Tokyo Toshiba Fuchu Corporation Inside the factory (72) Inventor Yoichi Inoue, No. 1, Toshiba-cho, Fuchu-shi, Tokyo Inside the Toshiba Fuchu factory

Claims (4)

[Claims] 1. An image from a camera that has captured the inside of an orbit is input to obtain a difference image from a predetermined initial image, a detection object is extracted from the difference image, and the detection object on the difference image is extracted. Calculation for estimating the distance from the position to the point where the camera and the detection object actually exist, and estimating the actual size of the detection object from this distance and the number of pixels of the detection object on the difference image. Means for comparing the actual size of the detected object estimated by the calculating means with a predetermined first threshold value to specify the type of the detected object; and the calculating means When the detected substance is extracted by, an alarm output unit that outputs an alarm, and a display unit that displays the type of the detected substance when the type of the detected substance is specified by the detected substance specifying unit In-orbit monitoring device. 2. The in-orbit monitoring device according to claim 1, wherein it is determined that a train is approaching when the number of pixels of the detection object extracted by the calculation means becomes equal to or larger than a second threshold value set in advance. When the train approach determining means and this train approach determining means determine that the train is approaching,
An in-orbit monitoring device having an image input stopping means for stopping the input of the image captured by the camera to the calculating means. 3. An arithmetic means for inputting an image from a camera, which is photographed on the platform, obtaining a differential image from a predetermined initial image, and extracting a detected object from the differential image, and an arithmetic means for extracting the detected object. Determination means for determining whether or not the detected object is present at the edge of the platform, and when the determination means determines that the detected object is present at the edge of the platform, a warning is issued by predicting the fall under the platform. A first prediction means for instructing the output of the first warning means, an alarm output means for outputting the warning in response to the warning output instruction of the first prediction means, and a prediction result of the first prediction means. A platform monitoring device having display means for displaying. 4. The platform monitoring device according to claim 3, wherein the distance from the position of the detected object on the difference image to the point where the camera and the detected object actually exist, which is provided in the calculation means, is calculated. Estimating means for estimating the actual size of the detected object from this distance and the number of pixels of the detected object on the difference image, and the actual size of the detected object estimated by this estimating means in advance. A detection object specifying unit that specifies the type of the detection object by comparing the determined first threshold value; and a measuring unit that measures the number of detection objects of the same type specified by the detection object specifying unit. When the number of the detected objects measured by the measuring means exceeds a predetermined number set in advance, the warning output means is instructed to output the alarm by predicting the fall under the platform. Platform monitoring device and a second prediction means for inputting the prediction result on the display means.