JP2021051464A - Intrusion detection device and intrusion detection method - Google Patents

Abstract

To accurately detect an intrusion of a worker into an intrusion restricted area even when a boundary of the intrusion restricted area is changed.SOLUTION: There is provided an intrusion detection device where a boundary L of an intrusion restricted area is divided by a movable marker 2 and which detects whether or not a worker W existing in a work area including the intrusion restricted area has intruded into the intrusion restricted area. The device includes: an image acquisition section 11A for acquiring a captured image obtained by imaging a work area; a worker identification section 11C for extracting an area of a worker W from a captured image acquired by the image acquisition section 11A with the use of a first learnt model 12A for recognizing the worker W; a boundary identification section 11B for identifying a position of the marker 2 so as to identify the boundary L from the position of the marker 2; and an intrusion determination section 11D for determining whether or not the worker W has intruded into the intrusion restricted area based on the boundary L to be identified by the boundary identification section 11B and an area of the worker W to be extracted by the worker identification section 11C.SELECTED DRAWING: Figure 1

Description

The present invention relates to a technique for monitoring the intrusion of workers into an intrusion restricted area where the position of a boundary is appropriately changed. The present invention is applied to, for example, an inspection line or a storage warehouse for goods having different dimensions, in which the boundary of an intrusion restriction area changes depending on the dimensions of the goods.

In the inspection line of manufactured products, for example, a part of the work area is set as an intrusion restriction area for the safety of workers until the product is transported and fixed to the inspection line. Then, after the safety of the worker is ensured, the worker inspects the product.
As a technique for defining an intrusion restriction area and preventing workers from entering the area, for example, there is a technique described in Patent Document 1. In Patent Document 1, when the intrusion detection device determines that the warning light is displaying the alarm color by analyzing the image and determines that the worker has invaded the intrusion restricted area, the entry is restricted. The area intruder detection system handles abnormal situations.

International Publication No. 2012/039277

The technique of Patent Document 1 is based on the premise that it is necessary to use an existing alarm and the intrusion restriction area is fixed.
However, the products manufactured in the steel industry have various dimensions depending on the products. For example, in a steel pipe inspection line, the range of the intrusion restriction area changes depending on the length of the steel pipe of the target product. It is also difficult to prepare a fixed alarm due to changes in the intrusion restriction area.
Further, since the workers who prepare for the inspection and perform the inspection take various postures and change their clothes depending on the season, there is a problem that the workers cannot be recognized by ordinary techniques.

An object of the present invention is to accurately detect the intrusion of a worker into the intrusion restriction area even when the intrusion restriction area changes appropriately depending on the product dimensions and the like.

The present inventor has found that it is not possible to accurately recognize a worker by diverting existing human recognition data that is generally used, and has obtained the finding that individual machine learning for recognizing a worker is required.

Then, in order to solve the problem, according to one aspect of the present invention, the boundary of the intrusion restriction area is divided by the movable sign, and the worker existing in the work area including the intrusion restriction area is moved to the intrusion restriction area. An intrusion detection device that detects the presence or absence of intrusion, and the image acquisition unit uses an image acquisition unit that acquires an image captured by capturing the work area and a first trained model that recognizes a worker. A worker specifying part that extracts a worker's area from the acquired captured image, a boundary specifying part that specifies the position of the marker and specifies the boundary from the position of the sign, and a boundary specified by the boundary specifying part. The gist is to include an intrusion determination unit for determining whether or not a worker has invaded the intrusion restricted area based on the worker area extracted by the worker identification unit.

Further, according to another aspect of the present invention, the boundary of the intrusion restriction area is divided by the movable sign, and the presence or absence of invasion of the worker existing in the work area including the intrusion restriction area into the intrusion restriction area is determined. Using the first trained model that recognizes the worker, which is an intrusion detection method for detecting, the step of extracting the worker's area from the captured image obtained by capturing the work area and the position of the marker are specified. , A step of specifying the boundary from the position of the sign, and a step of determining whether or not a worker has invaded the intrusion restricted area based on the specified boundary and the extracted worker area. The gist is to have.

According to the aspect of the present invention, even if the intrusion restricted area where the intrusion of workers is not permitted changes appropriately due to a change in product dimensions or the like, the intrusion of workers into the intrusion restricted area is allowed. It is possible to detect inaccurately and reliably monitor the intrusion of workers into the intrusion restricted area. As a result, according to the aspect of the present invention, for example, when a worker invades the intrusion restricted area, it is possible to promptly issue an alarm.
Here, even if the worker takes various working postures, the worker is recognized by the learning model pre-learned by machine learning such as deep learning, so that it is possible to surely recognize the area where the worker exists. is there.

It is a conceptual diagram for demonstrating the intrusion detection apparatus which concerns on embodiment based on this invention. It is a figure explaining the structure of the monitoring control unit which concerns on embodiment based on this invention. It is a conceptual diagram about the difference in product dimensions. It is a conceptual diagram which shows the example which arranges a plurality of image acquisition parts. It is a figure which shows the image example of the worker for generating the first trained model.

Next, an embodiment of the present invention will be described with reference to the drawings.
Here, the drawings are schematic, and the size and length ratio of each part are different from the actual ones. Further, the embodiments shown below exemplify a configuration for embodying the technical idea of the present invention, and the technical idea of the present invention is defined by the claims described in the claims. Various changes can be made within the technical scope of the invention.

(Constitution)
In the present embodiment, a steel pipe 1 as an example of a product manufactured in the steel industry will be illustrated, and a case where the intrusion detection device of the present embodiment is applied to the inspection line of the steel pipe 1 will be described as an example.
FIG. 1 is a conceptual diagram of an inspection line of a steel pipe 1 and an intrusion detection device.
In the inspection line of the steel pipe 1, the steel pipe 1 to be inspected is conveyed to the work area ARA-1 by the transfer device 14 (not shown), and when the inspection of the conveyed steel pipe 1 is completed (state in FIG. 1), the steel pipe 1 The steel pipe 1 is transported by driving the transport device 14 in the direction of arrangement. Then, when the steel pipe 1 is transported to the work area ARA-1 and when the transport device 14 is driven to transport the steel pipe 1 from the work area ARA-1, the installation position of the steel pipe 1 and the steel pipe 1 The vicinity of the area and the movable area of the steel pipe 1 are the intrusion restriction area ARA-2. Here, the transport device 14 is composed of, for example, a kicker device.

At this time, a boundary display sign 2 made of a signboard is provided above the position of the boundary L so that the worker W can recognize the movable boundary L of the intrusion restriction area ARA-2. The sign 2 is suspended from a wire rope (not shown) erected on the ceiling side. The wire rope is erected along the longitudinal direction of the steel pipe 1 placed on the work area ARA-1. Along the erected wire rope, the sign 2 can move back and forth in the longitudinal direction of the steel pipe 1 placed on the work area ARA-1. The movement of the sign 2 may be performed by remote control, or the sign 2 may be manually operated.
The sign 2 may be in any form as long as it can be moved and its position can be specified.

As shown in FIG. 3, the dimensions of the steel pipe 1 to be inspected are various. Currently, a sign 2 is set above the floor in front of the end of the steel pipe 1 having the maximum size among the steel pipes 1 transported to the work area ARA-1 for inspection. That is, when viewed from above, there is a buffer space between the position of the marker 2 and the end of the steel pipe 1.
In the present embodiment, the boundary of the intrusion restriction area ARA-2 is variable only on the sign 2 side. In the example of FIG. 1, the invasion of the worker W into the intrusion restriction area ARA-2 is considered to be from the left boundary L (mark 2 side) and the right boundary in FIG. In the present embodiment, the intrusion of the worker W from the boundary L (mark 2 side) on the left side in FIG. 1 is monitored.
Further, the movable boundary L may be 2 or more. A sign 2 that moves in the moving direction of the boundary L may be installed for each movable boundary L.
As shown in FIG. 1, the intrusion detection device of the present embodiment includes a surveillance camera 10, a surveillance control unit 11, and a storage unit 12.

<Surveillance camera 10>
The surveillance camera 10 images the work area ARA-1 and acquires the captured image. The imaging area of the surveillance camera 10 does not have to be the entire working area ARA-1, and the area in the working area ARA-1 near the movable boundary L (for example, an area around 1 m of the boundary L) can be imaged. Just do it. The surveillance camera 10 can overlook, for example, the intrusion restriction area ARA-2 and the workable area (the area other than the intrusion restriction area ARA-2 in the work area ARA-1), and the sign 2 at the boundary L between the two. Even if the height is high, the worker W, the sign 2 indicating the boundary L, the transport equipment, etc. should be installed at a height that allows a bird's-eye view.
The surveillance camera 10 executes imaging at a preset sampling cycle or at an appropriate timing.
The number of surveillance cameras 10 may be one. However, it is more desirable to have two or more surveillance cameras 10 installed at different positions. If there are two or more surveillance cameras 10, the positions of the worker W and the sign 2 can be converted into three-dimensional coordinates, and the positional relationship between the two W and 2 can be known more accurately.

<Monitoring control unit 11>
As shown in FIG. 2, the monitoring control unit 11 includes an image acquisition unit 11A, a boundary identification unit 11B, a worker identification unit 11C, and an intrusion determination unit 11D. Further, the monitoring control unit 11 of the present embodiment includes an alarm issuing unit 11E, a device stop unit 11F, and an alarm transmitting unit 11G.
Here, as shown in FIG. 1, the intrusion detection device of the present embodiment has an alarm 13 for notifying the intrusion of the worker W into the intrusion restriction area ARA-2. The alarm 13 may be provided at a position where the worker W can easily see it. The alarm 13 is installed, for example, in an area where the worker W approaching the end of the steel pipe 1 can see. In FIG. 1, the alarm 13 is shown on the left side, but if it is arranged on the right side of the steel pipe 1, it is easy for the worker W to see it.

Further, in the present embodiment, it is assumed that the alarm application 15A capable of outputting an alarm is installed on the mobile terminal 15 such as a smartphone owned by the worker W. It is assumed that the alarm application 15A is configured to start up and output an alarm when it receives an alarm transmission signal. The output of the alarm is, for example, an alarm sound.
Further, as shown in FIG. 2, the storage unit 12 stores the first trained model 12A and the second trained model 12B.

<First trained model 12A>
The first trained model 12A is generated as follows.
That is, it has various working postures, sizes, positions, and backgrounds as illustrated in FIG. 5 from a plurality of image data obtained by imaging the work area ARA-1 from a plurality of positions (preferably the installation position of the surveillance camera 10). A plurality of images of the worker W are cut out, and the cut out plurality of images are stored in the image database 20 as learning data (images). The number of images is, for example, 100,000.

Next, the image data stored in the image database is labeled (labeling process).
Next, using the labeled image data, learning is performed by a neural network or other machine learning to generate a first trained model 12A in which the feature of recognizing the worker W is extracted. It should be noted that each worker's working posture is individually learned by machine learning to generate two or more first trained models 12A that recognize the worker W, and two or more first trained models are generated. 12A may be used in the worker identification unit 11C.
As for the type of machine learning and the learning method for generating the first trained model 12A, a known method such as deep learning may be adopted.

<Second trained model 12B>
The second trained model 12B may also be generated by the same method as that of the first trained model 12A.
That is, images of the sign 2 captured from various places (positions, backgrounds) and directions are cut out from the imaged data at the site, and the cut out plurality of images are saved as learning data (images).
Next, the cut out image data is labeled.
Next, the labeled image data is trained by a neural network or other machine learning to generate a second trained model 12B from which the feature for recognizing the marker 2 is extracted.

<Image acquisition unit 11A>
The image acquisition unit 11A acquires the captured image captured by the surveillance camera 10.
<Boundary identification part 11B>
The boundary specifying unit 11B specifies the position of the sign 2, and from the specified position of the sign 2, the position of the movable boundary L of the intrusion restriction area ARA-2 is specified.
The boundary specifying unit 11B of the present embodiment uses the second learned model 12B that recognizes the sign 2 to perform a process of specifying the position of the sign 2 from the captured image acquired by the image acquisition unit 11A. That is, the target captured image is searched for a feature position similar to the feature by the second trained model 12B, and the position of the area where the marker 2 exists is extracted. The position of the sign 2 by the boundary specifying unit 11B needs to be specified only once each time it is determined that the movable boundary L has changed.

Then, the boundary specifying unit 11B sets the position below the area where the extracted sign 2 exists as the position of the boundary L of the intrusion restriction area ARA-2. The boundary L is a virtual line including the lower part of the center of gravity of the sign 2 and orthogonal to the advancing / retreating direction of the sign 2. The steel pipe 1 side of the boundary L is the intrusion restriction area ARA-2.
Here, the boundary specifying unit 11B may specify (estimate) the position of the marker 2 without using the second trained model 12B. For example, an indoor GPS may be installed on the sign 2 and the position of the sign 2 may be specified using the indoor GPS. Further, the position of the sign 2 may be specified according to the winding amount of the wire winding device for advancing and retreating the sign 2. Further, since the installation position of the steel pipe 1 is determined in advance, the dimensions of the steel pipe 1 to be tested are acquired from a host computer, and the position of the end of the steel pipe 1 to be installed in the inspection area is determined from the acquired length of the steel pipe 1. It may be estimated and the position of the marker 2 and the boundary L may be estimated.

<Worker identification unit 11C>
The worker identification unit 11C uses the first learned model 12A that recognizes the worker W to perform a process of extracting the region W-ARA of the worker W from the captured image acquired by the image acquisition unit 11A. That is, the target captured image is searched for a feature position similar to the feature by the first trained model 12A, and the region W-ARA in which the worker W exists is extracted.
The area W-ARA of the worker W is, for example, a quadrangular area.
Here, the captured image used by the boundary specifying unit 11B and the captured image used by the worker specifying unit 11C do not have to be the same. The captured image used by the boundary specifying unit 11B may be acquired once when the intrusion restriction area ARA-2 is installed, but the captured image used by the worker specifying unit 11C is in the intrusion restriction area ARA-2. During the setting, it is preferable to acquire at a predetermined sampling cycle and perform the worker extraction process. The sampling period is, for example, 0.5 seconds.

<Intrusion determination unit 11D>
The intrusion determination unit 11D invades the intrusion restriction area ARA-2 based on the boundary L specified by the boundary identification unit 11B and the area W-ARA of the worker W extracted by the worker identification unit 11C. Judge the presence or absence of.
For example, the intrusion determination unit 11D includes an area W-ARA in which the worker W extracted by the worker identification unit 11C exists and an area of the intrusion restriction area ARA-2 partitioned by the boundary L specified by the boundary identification unit 11B. If it is determined that there is an overlap, it is determined that the worker W has invaded the intrusion restriction area ARA-2. On the other hand, in other cases, the intrusion determination unit 11D determines that the worker W has not invaded the intrusion restriction area ARA-2.
The overlap may be determined by the overlap in a plan view, but it is determined only by the coordinate position in the advancing / retreating direction of the sign 2 (the direction orthogonal to the boundary L (referred to as the x direction)) out of the two horizontal directions. do it.

Since the recognition of the area W-ARA of the sign 2 and the worker W occupies a certain area on the image, the position may be specified by using the center of gravity point or the like. However, it is possible to detect intrusion at an early stage by making a judgment based on the overlap between the area W-ARA of the worker W and the intrusion restriction area ARA-2 (based on the coordinates of the boundary L position). When the determination is made based on the overlap, for example, even when only the arm of the worker W straddles the boundary L can be detected.
Further, the intrusion determination unit 11D may determine that there is an intrusion only when the worker W invades continuously for a predetermined time or more (for example, 1 second or more). In this case, it is possible to exclude the case where the worker W straddles the boundary L for a moment.

<Alarm alarm unit 11E>
When the intrusion determination unit 11D determines that the worker W has invaded, the alarm issuing unit 11E outputs an alarm command to the alarm device 13. The alarm device 13 notifies an alarm when an alarm command is input.
<Equipment stop 11F>
When the intrusion determination unit 11D determines that the worker W has invaded, the equipment stop unit 11F outputs a stop command to the equipment operating in the intrusion restriction area ARA-2. The device for which the stop command is input stops driving.

<Alarm transmitter 11G>
When the intrusion determination unit 11D determines that the worker W has invaded, the alarm transmission unit 11G transmits an alarm transmission signal to the alarm application 15A. The alarm application 15A to which the alarm transmission signal is transmitted performs a process of issuing an alarm. The transmission of the alarm transmission signal is executed, for example, via the Wifi device installed in the work area ARA-1, and is transmitted to the mobile terminal 15 owned by the worker W in the work area ARA-1.
Here, when one surveillance camera 10 captures the entire area of the work area ARA-1, there is a high possibility that the certification of the worker W will be inaccurate for the information at a position far from the surveillance camera 10. For example, a part of the worker W may be hidden by a steel pipe 1 or the like or equipment, and the recognition of the worker W may be inaccurate.

Further, depending on the positional relationship between the sign 2 and the worker W in the boundary L direction, the positions of the boundary L and the worker W appear to be inverted when viewed from the imaging direction of the surveillance camera located away from the sign. In some cases.
Therefore, in the present embodiment, as shown in FIG. 4, a plurality of surveillance cameras 10A to 10E are installed along the direction parallel to the moving direction (advancing / retreating direction) of the sign 2. Since the sign 2 is located above, it can be recognized without being hidden by the steel pipe 1, and therefore, the captured image of any surveillance camera 10 may be adopted.

When a plurality of surveillance cameras 10 are used as described above, the image acquisition unit 11A is captured by the surveillance camera 10 closest to the position of the sign 2 specified by the boundary identification unit 11B among the plurality of surveillance cameras 10. The captured image is acquired as an captured image for extracting the region W-ARA of the worker W, and is supplied to the worker specifying unit 11C.
Specifically, the work area ARA-1 is divided into a plurality of compartments Z1 to Z5 along the moving direction of the sign 2, and each of the compartments Z1 to Z5 is imaged by the surveillance cameras 10A to 10E toward the target compartment. The axes are arranged so that each surveillance camera 10 can capture at least the entire target division.

Here, as a method of determining the surveillance camera 10 closest to the position of the sign 2 only from the image of the surveillance camera 10, for example, the target section of each surveillance camera 10 is set in the horizontal direction of the imaging range of each surveillance camera 10. A method is conceivable in which a surveillance camera that is arranged in the center and has a marker 2 that moves to each surveillance image in the center of the horizontal direction of the imaging range is determined to be the surveillance camera that is closest to the marker 2.
Then, the image acquisition unit 11A supplies the image of the surveillance camera 10 in the category where the sign 2 is presumed to be located to the worker identification unit 11C.

By determining the positional relationship with the worker W using the image captured by the surveillance camera 10 at a position close to the sign 2, the positional relationship between the worker W and the sign 2 can be accurately determined on the image. It becomes.
The captured images of all the surveillance cameras 10A to 10E may be supplied to the worker specifying unit 11C, and all the captured images may be extracted by the worker.
In addition, based on the image captured by the camera 10 that captures the vicinity of the fixed boundary L on the right side in FIG. 1, the process of the worker identification unit 11C is executed, and an intruder from the right boundary of the intrusion restriction area ARA-2 is executed. Intrusion may be detected separately.

(Operation and others)
The intrusion detection device temporarily sets the area including the area on which the steel pipe 1 is placed in the work area ARA-1 as the intrusion restriction area ARA-2 when the steel pipe 1 is conveyed.
At this time, the sign 2 is moved back and forth to move the sign 2 to the vicinity of the end of the steel pipe 1. The movement of the sign 2 is performed as appropriate. When the work area ARA-1 is placed on the steel pipe 1, the sign 2 is set near the end of the steel pipe 1 at the time of inspection.
In the intrusion detection device, the surveillance camera 10 images the work area ARA-1, the boundary identification unit 11B identifies the position of the movable boundary L, and the worker identification unit 11C acquires the work area at a predetermined sampling cycle. Based on the captured image of ARA-1, the region W-ARA of the worker W is specified.

Then, the intrusion detection device determines whether or not the worker W has invaded the intrusion restriction area ARA-2 from the overlap of the specified boundary L position and the area W-ARA of the worker W. Specifically, the worker W determines the intrusion into the intrusion restriction area ARA-2 on the movable boundary L side.
When the intrusion detection device determines that the worker W is in the intrusion restriction area ARA-2, the intrusion detection device outputs an alarm signal to the alarm 13 to notify the intrusion from the alarm device 3. It will be issued.
At the same time, the intrusion detection device outputs an emergency stop command to the operating equipment (for example, the pipe transfer device 14) in the intrusion restriction area ARA-2, which is in danger of the worker W.
Further, the intrusion detection device transmits a signal instructing an alarm to notify that the worker W has directly entered the intrusion restriction area ARA-2 to the mobile terminal 15 (smartphone or the like). When the smartphone is instructed, the alarm application 15A starts up and issues an alarm.
It is not necessary to have all of the above alarm processing.

Here, when using the plurality of surveillance cameras 10, it is desirable to use the captured image of the surveillance camera 10 closest to the boundary L position specified by the position of the sign 2.
Further, by recognizing the same sign 2 in the captured images of two or more surveillance cameras 10 installed at different positions, it is possible to acquire the unified coordinate system of the sign 2 in the preset three-dimensional coordinate system. It becomes. In this case, the surveillance camera 10 closest to the sign 2 on the boundary L and the surveillance camera 10 existing on the opposite side of the sign 2 are selected and used for the recognition of the worker W. By first selecting the surveillance camera 10 closest to the sign 2, the determination accuracy when the worker W approaches the sign 2 is improved. Further, by using the surveillance camera 10 on the opposite side of the sign 2, the positions on both sides of the sign 2 can be recognized. At that time, it is more desirable to select the surveillance camera 10 farthest from the sign 2. This is because a wider monitoring area can be obtained.

In this way, when a plurality of surveillance cameras 10 are used at the same time, the worker W may recognize the unified three-dimensional coordinate system (the three-dimensional coordinates of the specific surveillance camera 10 may be unified). The position of the worker W can be confirmed by another unified three-dimensional coordinate system). Whether or not the worker W exists in the intrusion restriction area ARA-2 can be determined from the positional relationship between the recognized sign 2 positions and the recognized worker W.
As described above, the intrusion detection device of the present embodiment determines whether or not the worker W is in the intrusion restriction area ARA-2 in which the intrusion is restricted for safety in the factory or the like. When it is determined that there is a worker in the intrusion restriction area ARA-2, an alarm is issued to alert the worker W, and if a dangerous device is in operation, the device is stopped in an emergency, and the worker W's For safety. As a result, safe and stable factory operation can be realized, and safe and highly efficient factory operation can be realized.

That is, the intrusion detection device has an effect of being able to issue an alarm when the worker W enters the intrusion restriction area ARA-2.
Further, the intrusion detection device transmits an alarm to the mobile terminal 15 such as a smartphone owned by the worker W, so that the worker W is in a place where the alarm cannot be heard due to noise in the factory or the alarm lamp cannot be seen. However, the alarm application 15A of the smartphone owned by the worker W is directly activated, and the worker W can be alerted by notifying the danger.
In this way, the intrusion detection device can detect unsafe behavior of workers (intrusion into the operating line, etc.) by video analysis, issue an alarm or stop the equipment, and prevent occupational accidents. ..

1 Steel pipe 2 Sign 10 Surveillance camera 11 Surveillance control unit 11A Image acquisition unit 11B Boundary identification unit 11C Worker identification unit 11D Intrusion determination unit 11E Alarm alarm unit 11F Equipment stop unit 11G Alarm transmission unit 12 Storage unit 12A First learned Model 12B Second learned model 13 Alarm 14 Transport device 15 Mobile terminal 15A Alarm app ARA-1 Work area ARA-2 Intrusion restriction area L Boundary W Worker W-ARA Worker area

Claims (10)

An intrusion detection device in which the boundary of the intrusion restriction area is demarcated by a movable sign, and the presence or absence of intrusion of a worker existing in the work area including the intrusion restriction area into the intrusion restriction area is detected.
An image acquisition unit that acquires an image captured by capturing the above work area, and an image acquisition unit.
Using the first trained model that recognizes the worker, the worker identification unit that extracts the worker's area from the captured image acquired by the image acquisition unit, and the worker identification unit.
A boundary specifying part that identifies the position of the sign and specifies the boundary from the position of the sign,
An intrusion determination unit that determines whether or not a worker has invaded the intrusion restriction area based on the boundary specified by the boundary identification unit and the worker area extracted by the worker identification unit.
An intrusion detection device comprising. The intrusion detection device according to claim 1, wherein the first learned model is a model learned by machine learning based on a plurality of images of workers working in the work area. Claim 1 or claim 2 is characterized in that the boundary specifying unit identifies the position of the sign from the captured image acquired by the image acquisition unit using a second trained model that recognizes the sign. The intrusion detection device described in. The intrusion detection device according to claim 3, wherein the second trained model is a model learned by machine learning based on a plurality of images obtained by capturing the markers. The intrusion detection according to any one of claims 1 to 4, wherein the intrusion determination unit has an alarm issuing unit that outputs an alarm command to an alarm when it determines that a worker has invaded. apparatus. Any of claims 1 to 5, wherein the intrusion determination unit has a device stop unit that outputs a stop command to a device operating in the intrusion restriction area when it determines that a worker has invaded. The intrusion detection device according to item 1. The above sign is hung in a state where it can move forward and backward along a preset movement direction.
A plurality of imaging units are installed along a direction parallel to the above-mentioned moving direction.
Based on the position of the marker specified by the boundary specifying unit, the image acquisition unit obtains an image captured by the imaging unit closest to the boundary position among the plurality of imaging units by the worker specifying unit. The intrusion detection device according to any one of claims 1 to 6, wherein the image is acquired as an image for extracting an area of a worker. An alarm application that can output an alarm is installed on a mobile terminal that can be owned by a worker, and the alarm application is configured to start when an alarm transmission signal is received and output an alarm.
The intrusion according to any one of claims 1 to 7, wherein the intrusion determination unit has an alarm transmission unit that transmits an alarm transmission signal to the alarm application when it determines that a worker has invaded. Detection device. An intrusion detection method in which the boundary of the intrusion restriction area is demarcated by a movable sign, and the presence or absence of intrusion of a worker existing in the work area including the intrusion restriction area into the intrusion restriction area is detected.
Using the first trained model that recognizes the worker, a step of extracting the worker's area from the captured image obtained by capturing the work area, and
The step of identifying the position of the sign and identifying the boundary from the position of the sign,
Based on the above-specified boundary and the above-extracted worker area, a step of determining whether or not a worker has invaded the intrusion restricted area, and
An intrusion detection method comprising. The step of specifying the boundary is described in claim 9, wherein the position of the sign is specified from the captured image obtained by capturing the work area by using the second trained model that recognizes the sign. Intrusion detection method.

Images