JP2018067203A - Danger notification device, danger notification method, and calibration method for danger notification device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a danger notification device capable of more reliably determining a risk of a subject's falling from a bed and performing notification of the risk.SOLUTION: The danger notification device includes: distance image acquiring means for imaging a bed and a periphery of the bed from above the bed and acquiring a distance image by using a value of a distance to an object to be imaged as a pixel value; area setting means for setting, as a bed area, a bed top area extracted from the acquired distance image, and a correction area for preventing erroneous detection of a long-sitting state of a subject when the subject is on the bed; determination means for determining whether or not the subject is likely to fall from the bed based on the number of pixels of the subject outside the bed area in the distance image containing the subject; and notifying means for notifying at least one of the subject and a third party of a result of the determination from the determination means.SELECTED DRAWING: Figure 12C

Description

  The present invention relates to a danger notification device, a danger notification method, and a calibration method for the danger notification device.

  With the progress of an aging society in recent years, an accident of either a fall or a fall at a nursing facility or a hospital has occurred so much that it has become a big problem. For example, in nursing care facilities, there are many accidents of falling from a bed under the situation where care is not taken or watched over, and there may be a life-threatening risk if discovery is delayed.

For this reason, a bed removal detection device that detects bed movement may be used for a care recipient or patient who has been evaluated as having a high degree of risk based on the fall and fall assessment score sheet. Sensors used in the bed leaving detection device include, for example, a clip type attached to clothes, a mat type laid on the floor or bed, and the like, and detects a bed leaving operation of a care recipient.
However, in these sensors, there are many false detections, and in the case of the mat type, there is a problem that it is avoided across the mat.
In order to solve these problems, various bed leaving detection devices have recently been studied, and a device using a plurality of piezoelectric elements, a device using an ultrasonic sensor, a device using a distance image sensor, and the like have been proposed. (For example, refer nonpatent literature 1).

However, many of them detect a wake-up state or a leaving-bed state with a low risk of falling, so that the frequency of notifications is likely to increase, and it is not possible to detect only a fall from a bed with a high risk. In addition, there is a problem that it is difficult to detect a fall from a bedded state such as a slip-down from a bed with high risk.
For this reason, the danger alerting | reporting apparatus which can determine more reliably and can alert | report the danger of falling from a bed is desired.

Further, in the apparatus using the distance image sensor described in Non-Patent Document 1, the distance image sensor is installed on a pole or the like without being fixed to a wall or a ceiling, and is relatively easy to move. However, every time it is moved, calibration is required to identify the bed in the obtained distance image. As a calibration method for identifying a bed, a partially manual watershed method is used. Therefore, it is troublesome to perform manual calibration partly every time it is moved, and it may not be possible to accurately identify the bed. There is also the problem of being.
For this reason, the calibration method which can identify a bed correctly without an effort is desired.

Nakata Toyohisa, Kanai Hideaki, Kunifuji Susumu: "Bed fall detection system using distance information considering privacy", Japan Creativity Society, Proceedings of the 6th Knowledge Creation Support System Symposium, Graduate School of Knowledge Science, Japan Advanced Institute of Science and Technology February 26, 2009

  Therefore, the present invention aims to solve the above-described problems and achieve the following object. That is, an object of the present invention is to provide a risk notification device, a risk notification method, and a calibration method for the risk notification device that can more reliably determine and notify the danger of the subject falling from the bed. .

Means for solving the above-described problems are as follows. That is,
<1> Distance image acquisition means for capturing the image of the bed and the periphery of the bed from above and acquiring a distance image using the distance value to the object to be imaged as a pixel value; and the bed upper surface extracted from the acquired distance image An area setting means for setting, as a bed area, an area and a correction area for preventing erroneous detection of the long sitting position of the object person when the object person is on the bed; and the distance image including the object person Based on the number of pixels of the subject outside the bed area, a determination unit that determines whether or not there is a risk of the subject falling from the bed; A danger notification device comprising: notification means for notifying at least one of them.

In the danger notification device according to <1>, the distance image acquisition unit images the bed and the periphery of the bed from above the bed, and acquires a distance image using a distance value to the object to be imaged as a pixel value. . The region setting means is a correction for not erroneously detecting the upper surface area of the bed extracted from the distance image acquired by the distance image acquisition means and the long sitting state of the target person when the target person is on the bed. Set the area as the bed area. The determination unit determines whether or not there is a risk that the subject falls from the bed based on the number of pixels of the subject outside the bed region of the distance image including the subject. The notification means notifies the determination result of the determination means to at least one of the target person and a third party.
Accordingly, the subject who is informed that there is a risk of the subject falling from the bed moves so as not to protrude from the bed region, or the subject is moved by the third party to the bed region. In order not to protrude from the bed, the subject can be more reliably prevented from falling off the bed.

  <2> The area setting means further sets a bedside area for detecting the protrusion of the subject from the bed outside the bed area, and an area above the bed area and the bedside area. And whether or not there is a risk that the subject falls from the bed based on at least one of the number of pixels of the subject in the bedside region and the number of pixels of the subject in the upper region. The danger notification device according to <1> described above.

In the danger notification device according to <2>, the area setting unit includes a bedside area for detecting the protrusion of the subject from the bed outside the bed area, the bed area, and the bedside area. And an upper region of. The determination means determines whether or not there is a risk that the subject falls from the bed based on at least one of the number of pixels of the subject in the bedside region and the number of pixels of the subject in the upper region. To do.
Thereby, even when it is determined that a part of the subject exists outside the bed area, whether the subject is in the rising state on the bed or a part of the subject's body is the bed By determining whether it is in a state of protruding from the side, the risk of the subject falling from the bed can be more reliably determined and notified.

  <3> In the case where the number of pixels of the subject in the bedside region is equal to or greater than a predetermined number of pixels, the determination unit, when the number of pixels of the subject in the upper region is equal to or less than a predetermined number of pixels, The danger notification device according to <2>, in which it is determined that there is a risk that the subject falls from the bed.

In the danger notification device according to <3>, in the case where the number of pixels of the subject in the bedside region is equal to or greater than a predetermined number of pixels, the number of pixels of the subject in the upper region is predetermined. When the number of pixels is equal to or less than the number of pixels, it is determined that there is a risk that the subject falls from the bed.
Thereby, since it is possible to determine that a part of the subject's body protrudes from the bedside region and is unlikely to be in a wake-up state, the risk of the subject falling from the bed is more reliably determined, Can be notified.

  <4> The region setting unit further sets a region above the bed upper surface region in the upper region as the correction region, so that the determination unit does not erroneously detect the leaving state of the target person and a third party. The danger notification device according to any one of <2> to <3>.

  In the danger notification device according to <4>, the region setting unit further sets a region above the bed upper surface region in the upper region as the correction region, whereby the pixel of the subject is the bed upper surface region. Even if it exists in the area | region above the, it can determine that the said determination means does not misdetect as the said subject being in the state of getting out of bed with little risk of falling from a bed. In addition, even if a third-party pixel is present in the region above the bed upper surface region, the determination unit determines that the risk of the target person falling from the bed is low, so that the determination unit does not detect it erroneously. be able to. Thus, if it can determine so that the said determination means may not misdetect, it can determine more reliably and can alert | report the danger that the said subject will fall from a bed.

  <5> A distance image obtaining step of photographing the bed and the periphery of the bed from above the bed, and obtaining a distance image using a distance value to the photographing object as a pixel value, and a bed upper surface extracted from the obtained distance image A region setting step for setting, as a bed region, a region and a correction region for preventing erroneous detection of the subject's long-sitting position when the subject is on the bed; and the distance image including the subject Based on the number of pixels of the subject outside the bed area, a determination step for determining whether or not there is a risk of the subject falling from the bed, and a determination result by the determination step are obtained by the subject and a third party. And a notification step of notifying at least one of them.

In the danger notification method according to <5>, in the distance image acquisition step, the bed and the periphery of the bed are photographed from above the bed, and a distance image is obtained using the distance value to the photographing object as a pixel value. . In the region setting step, a bed upper surface region extracted from the acquired distance image and a correction region for preventing erroneous detection of the long sitting state of the subject when the subject is on the bed are used as the bed region. Set. In the determination step, based on the number of pixels of the subject outside the bed area of the distance image including the subject, it is determined whether there is a risk that the subject falls from the bed. In the notification step, the determination result in the determination step is notified to at least one of the target person and a third party.
Accordingly, the subject who is informed that there is a risk of the subject falling from the bed moves so as not to protrude from the bed region, or the subject is moved by the third party to the bed region. In order not to protrude from the bed, the subject can be more reliably prevented from falling off the bed.

  <6> Distance image acquisition means for capturing the distance image from the upper side of the bed and the periphery of the bed and acquiring a distance image using the distance value to the object to be imaged as a pixel value; and the bed upper surface extracted from the acquired distance image A method for calibrating a danger notification device having a region setting unit, and generating a gradient image obtained by dividing the acquired distance image into regions having similar spatial gradients, and based on the gradient image A method for calibrating a danger notification device, comprising: a bed upper surface area extracting step for extracting the bed upper surface area.

  In the calibration method of the danger notification device according to <6>, in the bed upper surface region extraction step, the acquired distance image is divided into regions having similar spatial gradients, whereby the upper surface of the bed, the floor, and the wall Therefore, the bed upper surface area can be extracted from the identified area.

  <7> The bed upper surface region extracting step further generates an edge image drawn along a portion where the change amount of the pixel value of the distance image is large, and based on a composite image of the edge image and the gradient image It is the calibration method for the danger notification device according to <6>, including a process of extracting the bed upper surface area from the divided area.

  In the calibration method of the danger notification device according to <7>, for example, when it is difficult to divide the upper surface of the bed and the floor where the spatial gradient is horizontal, based only on the gradient image, From the distance image, generate the edge image including a contour line drawn along a portion where the amount of change in the pixel value generated due to a difference in height between the upper surface of the bed and the floor is generated, and the floor image with respect to the floor By synthesizing the edge image including the contour line on the upper surface of the bed with the gradient image, the bed upper surface region can be more reliably extracted.

  <8> In the distance image acquired by the distance image acquisition means, among the number of pixels in the divided area, the distance image so that the number of pixels in the area corresponding to the upper surface of the bed is in a predetermined order. The risk notification device calibration method according to <7>, further including an installation step of installing an acquisition unit, wherein the bed upper surface region extraction step includes a process of extracting the predetermined rank region as the bed upper surface region. is there.

  In the method for calibrating a danger notification device according to <8>, for example, the distance image acquisition unit is installed so that the upper surface of the bed is extracted most widely in the distance image to be acquired, and the predetermined order is 1 When the position is set, the bed upper surface area can be automatically extracted from the distance image, so that the bed can be accurately identified without the need for calibration.

  <9> The danger notification device calibration method according to any one of <7> to <8>, further including a calculation step of calculating the position and orientation of the distance image acquisition unit based on the bed upper surface region.

  In the method for calibrating a danger notification device according to <9>, the position and position of the distance image acquisition unit are not manually set by calculating the position and orientation of the distance image acquisition unit based on the bed upper surface region. And the posture can be obtained, so that calibration can be performed without taking time and effort.

  <10> The danger notification device calibration method according to <9>, further including a region setting step of setting a region by dividing a space in the camera coordinate system based on the bed upper surface region.

  In the method for calibrating the danger notification device according to <10>, since the area is set by dividing the space in the camera coordinate system instead of the space in the world coordinate system that is susceptible to noise, Extraction can be made more reliably.

  <11> The danger notification device according to <10>, wherein the region setting step sets a correction region for preventing erroneous detection of the subject's long sitting position when the subject is on the bed. Calibration method.

  In the method for calibrating a danger notification device according to <11>, since the correction region for preventing erroneous detection of the subject's long-sitting state can be set, the erroneous detection can be reduced. It can prevent more reliably falling from the bed.

  <12> In the above <11>, in which the region setting step further sets a region above the bed upper surface region in the upper region as a correction region for preventing false detection of the subject's bed leaving state and a third party. It is a calibration method of the danger alarm device described.

  In the method for calibrating a danger notification device according to <12>, the region above the bed upper surface region in the upper region is further set as a correction region for preventing erroneous detection of the bed leaving state of the subject and the third party. By setting, false detection can be reduced, so that the subject can be more reliably prevented from falling from the bed.

  <13> Distance image acquisition means for capturing an image of the bed and the periphery of the bed from above and acquiring a distance image using the distance value to the object to be imaged as a pixel value; and the bed upper surface extracted from the acquired distance image An area setting means for setting an area, and a calibration method for a danger notification device, wherein the area is set by dividing a space in a camera coordinate system based on the bed upper surface area, and among the set areas, A danger notification device calibration method comprising: a region setting step for setting a correction region for preventing erroneous detection of the long sitting position of the subject when the subject is on the bed.

  In the method for calibrating a danger notification device according to <13>, since the area is set by dividing the space in the camera coordinate system instead of the space in the world coordinate system that is susceptible to noise, Extraction can be made more reliably.

  <14> The danger notification device calibration method according to <13>, further including a calculation step of calculating the position and orientation of the distance image acquisition unit based on the bed upper surface region.

  In the method for calibrating a danger notification device according to <14>, by calculating the position and orientation of the distance image acquisition unit based on the bed upper surface region, the position of the distance image acquisition unit is not manually set. And the posture can be obtained, so that calibration can be performed without taking time and effort.

  ADVANTAGE OF THE INVENTION According to this invention, the said various problems in the past can be solved, the danger alerting | reporting apparatus, the danger alerting | reporting method, and the danger alerting | reporting apparatus which can determine more reliably and can alert | report the danger that a subject falls from a bed A calibration method can be provided.

FIG. 1A is a schematic diagram showing an example of the danger notification device of the present invention. FIG. 1B is a block diagram showing an example of the danger notification device of the present invention. FIG. 2 is an explanatory diagram illustrating an example of a distance image acquired by the distance image acquisition unit. FIG. 3 is an explanatory diagram showing an example of region setting in the world coordinate system. FIG. 4A is an explanatory diagram showing a gradient image generated from the distance image of FIG. FIG. 4B is an explanatory diagram illustrating an image obtained by correcting the color unevenness of the gradient image in FIG. 4A. FIG. 4C is an explanatory diagram illustrating an image obtained by extracting the bed upper surface area and the floor area from the gradient image after the color unevenness correction in FIG. 4B. FIG. 4D is an explanatory diagram illustrating an edge image generated from the distance image of FIG. FIG. 4E is an explanatory diagram illustrating a composite image obtained by combining the edge image of FIG. 4D with the image obtained by extracting the bed upper surface region and the floor region of FIG. 4C. FIG. 4F is an explanatory diagram illustrating an image obtained by extracting the bed upper surface area from the composite image in FIG. 4E. FIG. 5 is an explanatory diagram illustrating another example of the distance image acquired by the distance image acquisition unit. FIG. 6A is an explanatory diagram showing a gradient image generated from the distance image of FIG. FIG. 6B is an explanatory diagram illustrating an image obtained by correcting the color unevenness of the gradient image in FIG. 6A. 6C is an explanatory diagram illustrating an image obtained by extracting the bed upper surface area and the floor area from the gradient image after the color unevenness correction in FIG. 6B. FIG. 6D is an explanatory diagram illustrating an edge image generated from the distance image in FIG. 5. 6E is an explanatory diagram illustrating a composite image obtained by combining the edge image of FIG. 4D with the image obtained by extracting the bed upper surface area and the floor area of FIG. 6C. FIG. 6F is an explanatory diagram illustrating an image obtained by extracting the bed upper surface area from the composite image in FIG. 6E. FIG. 7 is an explanatory diagram for calculating the camera position in the world coordinate system. FIG. 8 is an explanatory diagram showing an example of a distance image converted into luminance according to the height of the camera coordinate system. FIG. 9 is an explanatory diagram illustrating an example of a distance image including a subject in a long sitting position. FIG. 10 is an explanatory diagram showing an example of region setting in the camera coordinate system. FIG. 11A is an explanatory diagram illustrating an example of an image of a bed area including a correction area. FIG. 11B is an explanatory diagram illustrating another example of an image of a bed area including a correction area. FIG. 12A is a flowchart showing an example of a flow for calibrating and operating the danger notification device of the present invention. FIG. 12B is a flowchart showing an example of the flow of the calibration method for the danger notification device of the present invention. FIG. 12C is a flowchart showing an example of a flow for operating the danger notification device of the present invention. FIG. 13 is an explanatory diagram illustrating a distance image in the first embodiment. FIG. 14 is an explanatory diagram illustrating a distance image in the second embodiment. FIG. 15 is an explanatory diagram illustrating a distance image in the third embodiment. FIG. 16 is an explanatory diagram illustrating a distance image in the fourth embodiment. FIG. 17A is an explanatory diagram illustrating a gradient image after color unevenness correction in the fifth embodiment. FIG. 17B is an explanatory diagram illustrating an image of the bed upper surface area extracted in the fifth embodiment. FIG. 18A is an explanatory diagram illustrating an image in which a reference point is designated on the distance image acquired in Comparative Example 1. 18B is an explanatory diagram showing an image of the bed upper surface area extracted in Comparative Example 1. FIG. FIG. 19A is an explanatory diagram illustrating a gradient image after color unevenness correction according to the sixth embodiment. FIG. 19B is an explanatory diagram illustrating an image of the bed upper surface area extracted in the sixth embodiment. FIG. 20A is an explanatory diagram showing an image in which a reference point is designated on the distance image acquired in Comparative Example 2. FIG. 20B is an explanatory diagram showing an image of the bed upper surface area extracted in Comparative Example 2. FIG. 21A is an explanatory diagram illustrating a gradient image after color unevenness correction in the seventh embodiment. FIG. 21B is an explanatory diagram illustrating an image of the bed upper surface area extracted in the seventh embodiment. FIG. 22A is an explanatory diagram showing an image in which a reference point is designated on the distance image acquired in Comparative Example 3. FIG. 22B is an explanatory diagram showing an image of the bed upper surface area extracted in Comparative Example 3.

(Danger notification device and danger notification method)
The danger notification device of the present invention includes a distance image acquisition unit, a region setting unit, a determination unit, and a notification unit, and further includes other units as necessary.
The danger notification method of the present invention includes a distance image acquisition step, a region setting step, a determination step, and a notification step, and further includes other steps as necessary.
Since the danger notification method of the present invention can be suitably performed by the danger notification device of the present invention, the details of the danger notification method of the present invention will be clarified through the description of the danger notification device of the present invention.

<Distance image acquisition means and distance image acquisition step>
The distance image acquisition unit is a unit that captures an image of the bed and the periphery of the bed from above the bed, and acquires a distance image using a distance value to the object to be imaged as a pixel value.
The distance image obtaining step is a step of photographing the bed and the periphery of the bed from above the bed and obtaining a distance image using the distance value to the photographing object as a pixel value.
The distance image acquisition step can be suitably performed by the distance image acquisition means.

  The object to be photographed is only the bed and the periphery of the bed when calibrated by the calibration method of the danger notification device of the present invention described later, and the bed when the danger notification device of the present invention is operated. As long as the subject is included at least in addition to the periphery of the bed, there is no particular limitation, and it can be appropriately selected according to the purpose.

  The target person is not particularly limited as long as it is watched so as not to fall from the bed, and can be appropriately selected according to the purpose. For example, a person requiring dementia in a care facility, walking in a hospital Examples include patients with disabilities.

There is no restriction | limiting in particular as said bed, Although it can select suitably according to the objective, It is preferable that an upper surface is flat. If the upper surface of the bed is flat, it is advantageous in that it is easy to extract the region of the upper surface of the bed when calibrating.
There is no restriction | limiting in particular as a shape of the upper surface of the said bed, Although it can select suitably according to the objective, A rectangle is preferable.
There is no restriction | limiting in particular as inclination of the upper surface of the said bed, Although it can select suitably according to the objective, It is preferable that it is horizontal.

  The bed installation position is not particularly limited and may be appropriately selected according to the purpose. However, when the distance image acquisition unit is installed at a relatively low position and installed from the bedside direction, the distance image A position where the bedside (long side and short side of the bed) far from the installation position of the acquisition means can be in close contact with a wall or the like so that the subject does not fall down is preferable. When it cannot adhere to a wall or the like, it is preferable to provide a fence on the bed. The bedside far from the installation position of the distance image acquisition means becomes a blind spot and is difficult to measure, but this is advantageous in that it makes it easier to more reliably determine the risk of the subject falling from the bed. It is.

The distance image acquisition means is referred to as “distance image sensor”, “three-dimensional distance sensor”, “depth sensor”, “distance image camera”, and the like, for example, the distance to the subject to be photographed using infrared rays or the like. It is a three-dimensional measuring instrument that measures and images without contact. Specifically, the distance image acquisition means (hereinafter also referred to as “camera”) irradiates the imaging object with infrared rays, and takes time until the imaging element receives infrared rays reflected from the imaging object. Or a method based on the principle of triangulation from the parallax when the imaging device receives the infrared pattern reflected from the imaging object and irradiating the imaging object with an infrared pattern. The distance image can be acquired by obtaining the distance value up to and converting the obtained distance value into luminance as the pixel value.
In addition, the location where the imaging object does not reflect infrared rays, the location where reflected light is weakened even if infrared rays are reflected, the location where the shadow (occlusion) of the imaging object where the infrared pattern does not reach, etc. are the distance This is a missing area where no value can be obtained.

  There is no restriction | limiting in particular as installation height of the said distance image acquisition means, Although it can select suitably according to the objective, It is preferable to install in the comparatively high position near a ceiling. When the distance image acquisition means is installed at a relatively high position near the ceiling, the blind spots are reduced, so that it is difficult to detect falsely, and the risk of the subject falling from the bed is more reliably determined and notified. be able to.

  There is no restriction | limiting in particular as an installation direction of the said distance image acquisition means, Although it can select suitably according to the objective, It is preferable to install on the long axis of a bed. If the installation direction of the distance image acquisition means is on the long axis of the bed, it is less likely to be erroneously detected because both bedside floors that are likely to fall from the bed can be photographed with fewer blind spots, and the object The risk of the person falling from the bed can be more reliably determined and notified.

  The installation location of the distance image acquisition means is not particularly limited and can be appropriately selected depending on the purpose.For example, a weight that can be stably installed on a tripod, ceiling, wall, or floor is placed below. The upper part of the pole which has, the upper part of the arm stand attached to the said bed, etc. are mentioned.

It is preferable that the distance image acquisition unit includes an image sensor in which pixels are arranged two-dimensionally.
Commercially available products of the distance image acquisition means include, for example, those using infrared rays, Xtion PRO, Xtion PRO Live (hereinafter, manufactured by ASUS), Kinect (registered trademark) (manufactured by Microsoft), Swiss Ranger SR3000, Swiss. Examples include Ranger SR4000 (manufactured by MESA), DepthSense (manufactured by SoftKinetic), D-Imager (manufactured by Panasonic Corporation), and the like. In addition, in a stereo system using two cameras, Bumblebee (manufactured by Point Gray Research) or the like can be given.

<Area setting means and area setting process>
The region setting means uses a bed upper surface region extracted from the acquired distance image and a correction region for preventing erroneous detection of the long sitting state of the subject when the subject is on the bed. It is a means for setting.
In the region setting step, a bed upper surface region extracted from the acquired distance image and a correction region for not erroneously detecting the long sitting state of the subject when the subject is on the bed are used as a bed region. It is a process of setting.
The region setting step can be suitably performed by the region setting means.
The region setting step is included in a calibration method for the danger notification device of the present invention, which will be described later.

The bed area is an area where the risk of the subject falling is low, and includes the bed upper surface area and the correction area.
The bed upper surface area is an area obtained by extracting the upper surface of the bed from the distance image, and is extracted as a plane. Thereby, even if the subject is in a long sitting position with a low risk of falling, a part of the subject's body may protrude from the bed upper surface region. Therefore, the area setting means sets the correction area so that the determination means does not erroneously determine that the subject is in a long sitting position on the bed with a low risk of falling. To do. Note that the long sitting state means not only a posture in which a leg is extended long, but also a state in which the risk of falling from the bed is low.

Preferably, the area setting means further sets a bedside area for detecting the protrusion of the subject from the bed and an upper area in the bedside area and the bed area outside the bed area. . Accordingly, the determination unit can more reliably determine whether or not there is a risk that the subject falls from the bed based on the number of pixels of the subject in the bedside region and the number of pixels of the subject in the upper region. Can be determined. That is, even when it is determined that a part of the subject exists outside the bed area, whether the subject is in a safe wake-up state on the bed or a part of the subject's body is By discriminating and detecting whether it is a dangerous state that protrudes from the bedside, the risk of the subject falling from the bed can be more reliably determined and notified.
Note that the number of pixels of the subject means the number of pixels that can be identified from the subject.

  It is preferable that the area setting means further sets an area above the bed upper surface area in the upper area as the correction area. Thereby, even if the subject is in a bed leaving state with a low risk of falling or a state in which a part of the third party's body enters the bed upper surface region, the determination means is It is possible to determine so as not to misdetect the bed leaving state and the third party.

  Therefore, in order to more reliably determine and notify the danger of the subject falling from the bed, the area setting means sets the bed upper surface area and the correction area as bed areas, and the bed Outside the region, it is preferable to set the bedside region and the upper region. Further, the area setting means sets the bedside lower area in an area below the bedside area, and the determination means indicates that the subject is in an end sitting position where the subject sits on the bedside. It may be determined, and it may be determined whether or not the subject has fallen after the notification means notifies the danger.

  The upper area is an area above the bed upper surface area (hereinafter also referred to as “bed upper area”) and an area above the bedside area (hereinafter referred to as “bedside upper area”). It may be divided into a certain). Thereby, based on the number of pixels of the target person or the third party in another area, the third party can be detected in the bed upper area and the bedside upper area. In addition, the bed upper area may be used to detect the subject's getting up state or the third party in order to detect the subject's rising state.

  Next, based on the distance image acquired without photographing the target person and the third party, the area setting means, in the order of the bed upper surface area, the upper area, the correction area, and the bedside area. A process for extracting or setting each area will be described.

[Bed upper surface area extraction processing]
The bed upper surface region can be extracted based on the distance image acquired by the distance image acquisition unit as follows.

-Conversion from distance image coordinate system to camera coordinate system-
First, the distance image coordinates (u, v), which are the coordinates of the pixels in the distance image _, are converted into camera coordinates _{Pu, v} by the following equation (1).
The camera coordinate system with the origin of the viewpoint of the distance image acquisition means is represented by the X ₁ axis, the Y ₁ axis, and the Z ₁ axis that are orthogonal to each other. Further, X ₀ axis orthogonal world coordinate system representing the position of an object in a three-dimensional space, respectively, Y ₀ axis, and represented by Z ₀ axis.

However, in the above formula (1), k is the Z ₁ axial distance value per luminance in the distance image, I _{u, v} range image coordinates in the imaging elements of the distance image acquiring unit (u, v) The luminance of each pixel corresponding to, C _u and C _v are the optical axis center coordinates in the distance image coordinate system, a is the focal length f, and the physical distance s of the pixels in the imaging element of the distance image acquisition means It represents the ratio (s / f).
The focal length means the distance between the optical center (lens center) of the distance image acquisition means and the photographing projection plane.

Maximum distance D as the distance value k of Z ₁ axial per brightness is not particularly limited and may be appropriately selected depending on the purpose, for example, to measure the number of gradations h of luminance values of said distance image Therefore, k = D / h may be set. Specifically, when the distance image acquiring unit having a gradation number h of 256 is used to photograph a maximum of 3,200 mm away from the distance image acquiring unit, k = 12.5 mm / luminance is set.

  There is no restriction | limiting in particular as the maximum distance D, Although it can select suitably according to the objective, 2,000 mm or more and 10,000 mm or less are preferable, and 3,000 mm or more and 5,000 mm or less are more preferable.

The optical axis center coordinates (C _u , C _v ) are not particularly limited and can be appropriately selected according to the purpose. However, the optical axis center coordinates (C _u , C _v ) can be selected appropriately according to the specification of the distance image acquisition unit or the camera calibration method. In the case where the specification is not specified and the camera calibration method is not used, the center of the pixel constituting the distance image is preferable. Specifically, when the size of the distance image is 320 pixels × 240 pixels, it is preferable to set the optical axis center coordinates (C _u , C _v ) = (160, 120).

  The ratio a (s / f) is not particularly limited and may be appropriately selected according to the purpose. However, the specification of the distance image acquisition unit (focal length f and physical distance s of pixels), or It is preferable to obtain using a camera calibration method or the like. Even if the focal length f and the physical distance s of the pixels are not specified in the above specification, if the angle of view θ and the number of pixels w of the distance image are known, the following equation: a = (2 / w ) × tan (θ / 2), a can be calculated. Specifically, if the vertical image angle of the distance image acquisition unit is 45 degrees and the vertical pixel size of the distance image is 240 pixels, a = 0.34552.

-Generation of gradient images-
The inclination of the surface in the three-dimensional space can be represented by a third principal component that can be obtained by principal component analysis (PCA). From this, first, a set of camera coordinates P _{u, v} is generated by removing the missing region from the small section centered on the distance image coordinates (u, v), and the unit vector of the third principal component of the generated set N _{u, v} is obtained. However, the unit vector _{N u, v,} since there are two point target, the camera coordinate _{P u,} closer to _v, i.e., the camera coordinate _{P u, v} and the unit vector _{N u, v} inner product is positive And
There is no restriction | limiting in particular as said small division, According to the objective, it can select suitably, For example, if the size of the said distance image is 320 pixels x 240 pixels, it is good also as 11 pixels x 11 pixels.

Next, three components in the X ₁ axis direction, the Y ₁ axis direction, and the Z ₁ axis direction in the camera coordinate system of the obtained unit vector _{Nu, v} are converted into RGB (Red, Green, By substituting each element of the color of Blue), a gradient image color-coded according to the direction of the unit vector _{Nu, v} is generated.
The method of color-grading the gradient image is not particularly limited and may be appropriately selected depending on the purpose. For example, the elements of R _{u, v} , G _{u, v} , B _{u, v} may be interchanged.

-Color unevenness reduction correction for gradient images-
Since the generated gradient image may have a large amount of color unevenness due to the influence of noise or the like, it is preferable that correction for reducing the color unevenness is performed.
The correction for reducing the color unevenness of the gradient image is not particularly limited and can be appropriately selected according to the purpose. For example, correction for performing clustering on the obtained unit vectors _{Nu, v of} each pixel is available. Can be mentioned. In the clustering, color unevenness can be reduced by reducing the color so that the pixels that the unit vectors _{Nu, v} are approximated to be the same.

There is no restriction | limiting in particular as said clustering, Although it can select suitably according to the objective, For example, the k-means method etc. are mentioned.
The number of divisions (number of clusters) T3 used in the k-means method is not particularly limited and may be appropriately selected according to the purpose. However, the standard deviation from the representative point of each region is a predetermined threshold value or less. It is preferable to gradually increase so as to be determined.

  In the gradient image that has been subjected to the color unevenness reduction correction, when the upper surface of the bed and the floor surface on which the bed is installed are horizontal, the spatial gradient is the same. The floor is often the same color. Thereby, there is a possibility that the bed upper surface area and the floor area cannot be reliably divided. Therefore, before extracting the bed upper surface area and the floor area from the distance image as described below, or after extracting the bed upper surface area and the floor area, the amount of change in the pixel value of the distance image is as follows. It is preferable to generate an edge image drawn along a large portion, generate a composite image of the corrected gradient image and the edge image, and extract the bed upper surface region more reliably.

-Extraction of bed upper surface area and floor area-
Since the distance image acquisition means is installed so that the bed and the periphery of the bed are shown large, the area (number of pixels) of each region in the distance image divided by the clustering is compared. In many cases, the area of the floor region becomes large. For this reason, the bed upper surface region and the floor region are extracted by selecting a region having a large area among the regions divided in the distance image.

-Generation of edge images-
In the image obtained by extracting the bed upper surface region and the floor region, in order to ensure the division between the bed upper surface region and the floor region, a line is drawn along a portion where the change amount of the pixel value of the distance image is large. It is preferable to extract an edge and further generate an edge image from the distance image.
In the edge image, it is preferable that the edge obtained by drawing a line is black (for example, luminance value 0), and other than the edge is white (for example, luminance value 255).

  The edge extraction method is not particularly limited and may be appropriately selected depending on the purpose. For example, a canny edge detection algorithm may be used, and a line between the bed upper surface area and the floor area may be seamlessly displayed. Is preferably subtracted.

-Generation of composite image-
Next, a synthesized image is generated by synthesizing the edge image and the gradient image represented by the edge drawn along a step having a predetermined size, that is, a location where the change amount of the pixel value of the distance image is large. To do. Thereby, even if the upper surface of the bed and the floor surface on which the bed is installed are horizontal and the bed upper surface region and the floor region may not be reliably divided, the gradient image, and By synthesizing the edge image that can clarify the level difference between the bed upper surface area and the floor area, the bed upper surface area can be extracted more reliably.

  The generation of the composite image is not particularly limited and may be appropriately selected according to the purpose. However, the edge of the edge image is black (for example, luminance value 0), and other than the edge is white (for example, luminance). If the value is 255, it is preferable to multiply the luminance value of each corresponding pixel in the gradient image and the edge image. As a result, a binarized composite image can be generated by superimposing a region having a step and a region having a surface angle difference.

-Extraction of bed upper surface area-
The distance image acquisition means is installed in advance so that the number of pixels in the area corresponding to the upper surface of the bed has a predetermined order among the number of pixels in the divided area, Extracted as the bed upper surface area.
Specifically, the distance image acquisition unit is installed in advance so that the number of pixels in the area corresponding to the upper surface of the bed is maximized, and the composite image, or the bed upper surface area and the floor area are extracted. A labeling process for giving the same identification number to each pixel in the same region is performed on the image, and a label (region) having the maximum number of pixels is extracted as the bed upper surface region. In addition, among the number of pixels in the divided area, the distance image acquisition unit is installed in advance so that the number of pixels in the area corresponding to the upper surface of the bed is second, and the second area is set as the bed upper surface area. You may make it extract.

[Upper area setting process]
The upper region can be set, for example, by calculating the posture and position of the camera as follows.

-Calculation of camera posture-
The camera posture is expressed as the inclination of the distance image acquisition means in the world coordinate system, that is, the inclination N of the bed or floor in the camera coordinate system.
N may be a representative vector of the maximum class of clustering performed when extracting the bed upper surface area and the floor area (a unit vector that minimizes the sum of squares of distances when the k-means method is used). However, an image obtained by extracting the bed upper surface area and the floor area may include noise of peripheral objects such as a desk.
Therefore, in order to reduce the influence of noise as much as possible, the distance value I _{u, v} of the distance image with respect to the pixel (u, v) on the bed upper surface region is obtained again, and the camera coordinates P _u, Convert to _v and perform principal component analysis for each subdivision. Next, a unit vector N _{u, v} of the third principal component of the principal component analysis is obtained. However, in the same manner as described above, the unit vector _{N u, v,} since there are two point target, the camera coordinate _{P u,} closer to _v, i.e., the camera coordinate _{P u, v} and the unit vector _{N u, v} of The inner product is positive.
A unit vector that minimizes the sum of squares of the distances from the unit vectors N _{u, v} is defined as the gradient N of the distance image acquisition means.

-Calculation of camera position-
First, a height l in the world coordinate system from the distance image acquisition means to the upper surface of the bed and a height m in the world coordinate system from the distance image acquisition means to the floor surface are calculated.
Let P _i be a vector from the distance image acquisition means to an arbitrary point on the upper surface of the bed, a height l _i in the world coordinate system from the distance image acquisition means to an arbitrary point on the upper surface of the bed, that is, a vector _{Z 0} axis component of the P _i can be calculated by the following equation (3).

Further, if a vector from the distance image acquisition unit to an arbitrary point on the floor surface is Q _i , a height m _i in the world coordinate system from the distance image acquisition unit to an arbitrary point on the floor surface, that is, a vector The Z ₀ axis component of Q _i can be calculated by the following equation (4).

However, in the above formula (5), Average () represents an average value.

However, in Formula (6), Mode {} represents the mode value.

The equation to calculate the height l _i by (3), calculates the average value l of height l _i by the following equation (5).
Moreover, to calculate the height _{m i} by the formula (4), calculates the mode m height _{m i} by the following equation (6).
The reason why the average value is calculated at the height l _{i is} that it is assumed that the distance image acquisition unit can measure the distance to the upper surface of the bed with high accuracy. Further, the in the range of the mode to calculate the height m _i, when measuring the distance to the floor, surrounding objects such as a desk in addition the floor is a noise by the distance image acquiring unit This is because there is a possibility, and the use of the average value makes it easy to be influenced.

-Setting the upper area-
In the distance image, the luminance corresponding to the distance value obtained by the distance image acquisition unit is set for each pixel. After calculating the camera coordinates P _{u, v according} to the above equation (1), I1 _{u, v} in the following equation (7) is obtained, whereby the brightness corresponding to the Z _0- axis component of the world coordinate system is expressed as a pixel. Can be set again.

However, in the above formula (7), g ₁ represents a brightness difference between the bed upper surface and the floor surface, g ₂ represents luminance of the floor surface.

The g _1, not particularly limited and may be appropriately selected depending on the intended purpose, but is preferably 20 or more and 100 or less, more preferably 40 or more 70 or less.
The g _2, not particularly limited and may be appropriately selected depending on the intended purpose, preferably 200 to 255, more preferably 240 or more 255 or less.

If the luminance I1 _{u, v} is less than I _T of formula (8), the distance image coordinates (u, v) is that they are present in the upper region. Therefore, if I1 u, _v is less than I _T, the distance image coordinates (u, v), determines that contained in said upper region in the distance image.

However, in said Formula (8), d represents the height from the upper surface of the said bed in a world coordinate system, and is a constant for dividing the said bed area | region, the said bedside area | region, and the said upper area | region.

  There is no restriction | limiting in particular as d, Although it can select suitably according to the objective, 300 mm or more and 700 mm or less are preferable, and 400 mm or more and 600 mm or less are more preferable.

Similarly, the luminance value calculated by the above formula (7) _{(I1 u, v)} is larger than _{I T2} of the formula (9), the distance image coordinate coordinates (u, v), the bedside lower region It will be included.

However, e represents the height of the bedside lower region from the floor surface in the world coordinate system, and is a constant for dividing the bedside region and the bedside lower region.

  e can be appropriately selected according to the purpose as long as it is not more than the height from the floor surface to the bed upper surface in the actual space, but is preferably 200 mm or more and 500 mm or less, more preferably 300 mm or more and 400 mm or less.

[Correction area setting process]
The modified region, for example, can be set by such relative to the total pixels of the bed top surface area as follows to move the upper surface of the bed virtually in Z ₀ axis direction.

First, among the RGB components of the distance image coordinates (u, v) in the correction area, the luminance value of the B component is represented by Ib _{u, v} , and the luminance value of the R component is represented by Ir _{u, v} .

(I) When the distance image coordinates (u, v) are within the bed upper surface region, Iru _{, v} and Ibu _{, v} are initialized as 255 and g ₂ -g ₁ , respectively. Otherwise, Ir _{u, v} and Ib _{u, v} are initialized as 0.

(Ii) One pixel is selected from the bed area (the selected coordinates are (u, v)), the luminance values I _{u, v} at the coordinates of the distance image are obtained, and the camera coordinates are obtained by the above equation (1). The values P _{u, v} are calculated.

(Iii) b the upper surface of the bed in virtually _{Z 0} axis direction (initially and b = 1.) Determining the distance image coordinates when you move (u2, v2). The camera coordinate values P2 _{u, v} = (P2 _x , P2 _y , P2 _z ) after movement can be calculated by the following expressions (10) and (11). The following equation (11) can be obtained by solving for u and v in the above equation (1).

(Iv) If the Z distance image value at the distance image coordinates (u2, v2) when moving virtually b in the Z ₀ axis direction is I2 _{u, v} , I2 _{u, v} is obtained from the following equation (12). Can be derived.

(V) The pixel value in the image of the correction area is updated by the following formula (13) and the following formula (14).

(Vi) Count b and repeat (iii) to (v) until b is about 1,800 mm, which is about the height of a person.

(Vii) Another pixel in the bed upper surface area is selected, and (ii) to (vi) are performed on all the pixels in the bed upper surface area.

[Setting process for each area]
The following procedure determines whether the coordinates (u, v) on the distance image are included in the bedside region, the bedside upper region, the bedside lower region, the bed upper region, or the bed region. To do.

(I) The distance image is converted into a Z distance image using the above equation (7). The converted luminance value is I1 _{u, v} .

(Ii) The region in the height direction of the distance image coordinates (u, v) is determined using the above formula (8) and the above formula (9). Specifically, it is determined that _{I1 u, v} is greater than _{I T2} of the formula (9), the coordinates (u, v) is present in the bedside lower region. In other cases, when I1 _{u, v} is smaller than I _{T in the} above formula (8), the process proceeds to (iii-a), and when I1 _{u, v} is equal to or greater than I _{T in the} above formula (8), Move to (iii-b).

(Iii-a) When I1 _{u, v} is equal to or less than R2 _{u, v and} equal to or greater than B2 _{u, v of the} correction area image, the distance image coordinates (u, v) are included in the bed upper area. judge. Otherwise, it is determined to be included in the bedside upper area.

(Iii-b) When I1 _{u, v} is equal to or less than R2 _{u, v and} equal to or greater than B2 _{u, v of the} correction area image, it is determined that the distance image coordinates (u, v) are included in the bed area. To do. In other cases, it is determined to be included in the bedside region.

[Counting the area of each area]
After determining which region the distance image coordinates (u, v) are included in all the pixels on the distance image by the above method, the number of pixels (area) included in each region is counted. Here, the case where it is included in either the bedside upper region or the bed upper region is determined as the upper region, and the number of pixels in the upper region is defined as AII _-III . Further, the number of pixels of the bed-side region and A _IV.

  Thus, it is preferable to set each area with the distance image coordinates (u, v) without converting the obtained distance value from the camera coordinate system to the world coordinate system. Accordingly, it is possible to suppress a decrease in drawing performance, an increase in calculation amount, and an occurrence of occlusion due to the generation of many blank points due to sampling due to the conversion to the world coordinate system.

Further, the pixel values of the upper region A _II-III and the bedside region A _IV obtained as described above are set as initial values, and the pixel values are not less than a predetermined range in each region during operation of the danger notification device of the present invention. The determination means determines a state in which the risk of falling from the bed is high, with the changed number of pixels as the number of pixels of the subject.

<Determination means and determination process>
The determination unit is a unit that determines whether or not there is a risk that the subject falls from the bed based on the number of pixels of the subject outside the bed region of the distance image including the subject.
The determination step is a step of determining whether or not there is a risk that the subject falls from the bed based on the number of pixels of the subject outside the bed region of the distance image including the subject.
The determination step can be suitably performed by the determination unit.

  When the number of pixels of the subject in the bedside area is equal to or greater than a predetermined number of pixels, the determination means, when the number of pixels of the subject in the upper area is equal to or less than a predetermined number of pixels, It is preferable to determine that there is a risk of falling from the bed. Thereby, since it is possible to determine that a part of the subject's body protrudes from the bedside and is less likely to be in a wake-up state, the risk of the subject falling from the bed is more reliably determined, Can be notified.

This is because when the subject gets up, the number of pixels of the subject in the bedside region increases and the number of pixels of the subject in the upper region also increases. Also, when a visitor visits, the number of pixels of the subject in the bedside region increases and the number of pixels of the subject in the upper region also increases. By preventing the subject from determining that there is a risk of falling from the bed in these states, the accuracy of the determination can be increased.
Note that the determination that the subject has a risk of falling from the bed may be based on the fluctuation speed of the subject's pixels or based on a series of movements of the subject's pixels.

  Examples of the operation and state of the subject who is at risk of falling from the bed include a pick-up operation in a bedded state, an operation of a bed rail in a bedded state, and a pathological cause.

<Notification means and notification process>
The notifying means is means for notifying the determination result of the determining means to at least one of the target person and the third party.
The notification step is a step of notifying the determination result of the determination step to at least one of the target person and the third party.
The notification step can be suitably performed by the notification unit.

  The notification means is not particularly limited as long as the determination result of the determination means can be notified to at least one of the target person and the third party, and can be appropriately selected according to the purpose. Examples include display means, acoustic means such as an alarm sound, means for turning on illumination, means for imparting vibration, or a combination of these means.

  Examples of the third party include a visitor who has visited the subject, a caregiver at a care facility, a medical staff at a hospital, and the like.

<Other means and other processes>
There is no restriction | limiting in particular as said other means, According to the objective, it can select suitably, For example, a subject person protection means, a control means, etc. are mentioned.
There is no restriction | limiting in particular as said other process, According to the objective, it can select suitably, For example, a subject protection process, a control process, etc. are mentioned.
The other steps can be suitably performed by the other means.

The target person protection means is not particularly limited and may be appropriately selected according to the purpose. For example, when the determination means determines that the target person has a risk of falling from the bed, the target person It is a means that can be protected even if a person falls from the bed, such as an air bag.
There is no restriction | limiting in particular as said subject person protection process, According to the objective, it can select suitably, For example, when it determines with the said subject having the risk of falling from the said bed by the said determination process, an airbag The process etc. which can be protected even if the said subject falls by using etc. are mentioned.
The subject person protection step can be suitably performed by the subject person protection means.

The control means includes a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), and the like that control each operation of the danger notification device of the present invention, and controls the operation of the entire danger notification device. Various processes are executed on the basis of a control program.
The said control process is a process of performing various processes based on the control program for controlling the operation | movement of the whole danger notification apparatus, and can be suitably performed by the said control means.

(Calibration method of danger alarm device)
The first embodiment of the calibration method of the danger notification device of the present invention includes a bed upper surface region extraction step, preferably includes an installation step, a calculation step, and a region setting step, and other if necessary. These steps are included.

<Bed upper surface area extraction process>
The bed upper surface region extracting step is a step of generating a gradient image obtained by dividing the acquired distance image into regions having similar spatial gradients, and extracting the bed upper surface region from the region divided based on the gradient image. is there. In the bed upper surface region extraction step, the obtained distance image is divided into regions having similar spatial gradients, so that each region such as the bed upper surface, floor, and wall can be identified by labeling or the like. The bed upper surface area can be extracted from the area.

<Installation process>
In the distance image acquired by the distance image acquisition unit, the installation step includes the number of pixels in the divided area, and the number of pixels in the area corresponding to the upper surface of the bed is in a predetermined order. This is a step of installing distance image acquisition means. As described above, when the distance image acquisition unit is installed so that the upper surface of the bed is extracted most widely, the bed upper surface region can be automatically extracted in the region setting step.

<Calculation process>
The calculation step is a step of calculating the position and orientation of the distance image acquisition unit based on the bed upper surface region.
In the calculation step, the position and orientation of the distance image acquisition unit can be obtained without manually setting by calculating the position and orientation of the distance image acquisition unit based on the bed upper surface area. It can be calibrated without applying.

<Region setting process>
The region setting step is a step of setting a region by dividing a space in the camera coordinate system based on the bed upper surface region.
As described above, in the region setting step, in addition to the bed upper surface region, the correction region is set as the bed region having a low risk of falling, thereby further ensuring the risk of the subject falling from the bed. Can be determined and notified. Further, since the area is set by dividing the space in the camera coordinate system instead of the space in the world coordinate system which is easily affected by noise, the bed upper surface area can be extracted more reliably.

As the region setting step, it is preferable to set a correction region for not detecting the long sitting state of the subject when the subject is on the bed, and above the bed upper surface region in the upper region. It is more preferable to further set this area as a correction area for not detecting the bed leaving state of the subject person and the third party.
By setting a correction region for not detecting the subject's long-sitting position, erroneous detection can be reduced, so that the subject can be more reliably prevented from falling from the bed. Moreover, since the area above the bed upper surface area in the upper area is further set as a correction area for not detecting the bed leaving state of the subject and the third party, false detection can be reduced. It is possible to more reliably prevent the subject from falling from the bed.

The second embodiment of the calibration method of the danger notification device of the present invention includes an area setting step, preferably includes a calculation step, and further includes other steps as necessary.
The region setting step in the present embodiment sets the region by dividing the space in the camera coordinate system based on the bed upper surface region, and when the target person is on the bed among the set regions, This is a step of setting a correction region for preventing erroneous detection of the subject's long sitting position.

Here, an example of the danger notification device according to the present invention will be described with reference to the drawings.
In addition, in each drawing, the same code | symbol is attached | subjected to the same component and the overlapping description may be abbreviate | omitted. In addition, the number, position, shape, and the like of the following constituent members are not limited to the present embodiment, and can be set to a preferable number, position, shape, and the like in practicing the present invention.

FIG. 1A is a schematic diagram illustrating an example of a danger notification device according to the present invention. The X ₀ axis, the Y ₀ axis, and the Z ₀ axis are orthogonal to the world coordinate system, and the X ₁ axis is orthogonal to the camera coordinate system. , Y ₁ axis and Z ₁ axis. FIG. 1B is a block diagram showing an example of the danger notification device of the present invention.
As shown in FIG. 1A and FIG. 1B, the danger notification device 100 captures a distance image by photographing the bed 30 and the periphery of the bed 30 from above the bed 30 and using the distance value to the photographing object as a pixel value. The bed area includes the acquisition means 10, the bed upper surface area of the bed 30 extracted from the acquired distance image, and the correction area for preventing erroneous detection of the long sitting position of the target person when the target person is on the bed 30. And setting means for determining whether there is a risk that the subject falls from the bed based on the number of pixels of the subject outside the bed region of the distance image including the subject. 12 and a notifying means 13 for notifying the determination result of the determining means to at least one of the target person and a third party.
Note that the area setting unit 11, the determination unit 12, the notification unit 13, and the control unit 14 are stored and operated as programs in a storage unit of an information processing apparatus 15 such as a personal computer (not shown).

In the danger notification device 100, the distance image acquisition means 10 (Xtion PRO, 320 pixels × 240 pixels, manufactured by ASUS) is attached to the tripod 20 so that it can move freely.
Distance image acquiring unit 10 is installed in a direction across the bed area objects appear, Z ₁ axis (optical axis) direction of the distance value in the camera coordinate system can be obtained, as shown in FIG. 2 on the basis of the distance value A distance image can be acquired.

FIG. 2 is an explanatory diagram illustrating an example of the distance image acquired by the distance image acquisition unit, where the horizontal axis is the U axis in the distance image, and the vertical axis is the V axis orthogonal to the U axis in the distance image. is there.
As shown in FIG. 2, the distance image is represented such that the smaller the distance value obtained by the distance image acquisition means 10, the smaller the luminance, and the larger the distance value, the larger the luminance. Note that the area in which the distance image is partially black is the missing area where the distance value could not be obtained.
The height from the floor on which the distance image acquisition means is installed is 2,460 mm from the above formula (6), and from the distance image acquisition means to the upper surface of the bed from the above formula (5). When the height of was found, it was 1,718 mm. Further, setting the d in FIG. 3 to 500 mm, it was 149.6 When seeking _{I T} from the above equation (8).

FIG. 3 is an explanatory diagram showing an example of area setting in the world coordinate system, and shows each area set in the X ₀ Z ₀ plane.
As shown in FIG. 3, the distance image acquisition unit 10 is installed in the vicinity of the ceiling 50 so that the bed 30 installed on the floor surface 40 and the periphery of the bed 30 can be photographed. In FIG. 3, a region I indicates the bed region, a region II and a region III indicate the bed upper region, a region IV indicates the bedside region, and a region V indicates the bedside lower region. Further, the region II indicates the bed upper region, and the region III indicates the bedside upper region.

FIG. 4A is an explanatory diagram showing a gradient image generated from the distance image of FIG.
As shown in FIG. 4A, the gradient image includes components in three directions of the X ₁ axis, the Y ₁ axis, and the Z ₁ axis of the unit vector _{Nu, v} obtained in each pixel as elements of RGB colors. By replacing, the color is generated according to the direction of the unit vector _{Nu, v} .

FIG. 4B is an explanatory diagram illustrating an image obtained by correcting the color unevenness of the gradient image in FIG. 4A.
Since the gradient image in FIG. 4A has a large amount of color unevenness _, by clustering the unit vectors _{Nu, v} of each pixel by the k-means method, as shown in FIG. It is possible to obtain the gradient image divided into flat areas without color unevenness. In the gradient image subjected to the color unevenness reduction correction, the upper surface of the bed and the floor surface are horizontal, and the spatial gradient is the same, so the same color is obtained.

FIG. 4C is an explanatory diagram illustrating an image obtained by extracting the bed upper surface area and the floor area from the gradient image after the color unevenness correction in FIG. 4B.
When the gradient image of the color unevenness correction in FIG. 4B is subjected to a labeling process for giving the same identification number to each pixel in the same region and an area having a predetermined number of pixels or more is extracted, as shown in FIG. The floor area is extracted.

FIG. 4D is an explanatory diagram illustrating an edge image generated from the distance image of FIG.
As shown in FIG. 4D, the edge image of the distance image can be obtained by using Canny's edge detection algorithm for the distance image of FIG.

FIG. 4E is an explanatory diagram illustrating a composite image obtained by combining the edge image of FIG. 4D with the image obtained by extracting the bed upper surface region and the floor region of FIG. 4C.
As shown in FIG. 4E, the bed upper surface region and the floor region can be more reliably divided by synthesizing the edge image with the image obtained by extracting the bed upper surface region and the floor region.

FIG. 4F is an explanatory diagram illustrating an image obtained by extracting the bed upper surface area from the composite image in FIG. 4E.
The composite image in FIG. 4E is labeled, and a label having the maximum number of pixels is extracted as the bed upper surface area as shown in FIG. 4F.

FIG. 5 is an explanatory diagram illustrating another example of the distance image acquired by the distance image acquisition unit.
In FIG. 5, the height from the floor on which the distance image acquisition unit is installed is calculated as 1,680 mm from the above formula (6), and from the distance image acquisition unit, the height of the bed is calculated from the above formula (5). The height up to the upper surface was 1,108 mm. Further, setting the d in FIG. 3 to 500 mm, it was 149.6 When seeking _{I T} from the above equation (8).

6A to 6F extract the bed upper surface region from the distance image of FIG. 5 in the same manner as the images of FIGS. 4A to 4F generated from the distance image of FIG. 2 and the distance image of FIG. FIG. 6 is an explanatory diagram showing the processes up to and including a gradient image, a color unevenness corrected image, an image obtained by extracting a bed upper surface area and a floor area, an edge image, a composite image, and the bed upper surface area.
As shown in FIG. 6C, in the image obtained by extracting the bed upper surface area and the floor area, the bed upper surface area and the floor area may be connected. However, by using the edge image of FIG. 6D, the bed upper surface area and the floor area are separated in FIG. 6E, and the bed upper surface area is accurately extracted as shown in FIG. 6F.

FIG. 7 is an explanatory diagram for calculating the camera position in the world coordinate system.
In order to automatically set the outside of the bed region by calculating the camera position, the height l in the world coordinate system from the distance image acquisition means 10 to the upper surface of the bed 30, and the distance image acquisition means 10 to the floor surface 40 The height m in the world coordinate system is calculated. As shown in FIG. 7, the vector from the distance image acquiring unit 10 to any point of the upper surface of the bed 30 when the P _i, distance world coordinate system from the image acquisition unit 10 to any point of the upper surface of the bed 30 The height l _i , that is, the Z ₀ axis component of the vector P _i can be calculated by the above equation (3).

Furthermore, the vector from the distance image acquiring unit 10 to an arbitrary point on the floor 40 when the Q _i, the height m _i in the world coordinate system from the distance image acquiring unit 10 to any point of the floor _40, i.e. the vector The Z ₀ axis component of Q _i can be calculated by the above equation (4).
Using the above equation (3), the average value l of the height l _i is calculated by the above equation (5). Further, by using the above equation (4), calculates the mode m height m _i by the equation (6).

FIG. 8 is an explanatory diagram showing an example of a distance image converted into luminance according to the height of the camera coordinate system.
For example, the distance image with luminance corresponding to the distance values as shown in FIG. 2, as shown in FIG. 8, to be the Z ₀ distance image converted to a luminance corresponding to the height of the camera coordinate system it can.

FIG. 9 is an explanatory diagram illustrating an example of a distance image including a subject in a long sitting position.
As shown in FIG. 9, in the distance image including the subject in the long sitting position, even if the subject is in a state where the risk of falling is small, a part (enclosed portion) of the subject's body is present. Since it is outside the bed upper surface area, it is determined that the danger is high. For this reason, in the danger notification device of the present invention, for the convenience of the area setting in the camera coordinate system as shown in FIG. 10, the image is taken as the bed area, but there is actually an area outside the bed area.

FIG. 10 is an explanatory diagram showing an example of region setting in the camera coordinate system.
As shown in FIG. 10, as the correction area, it is necessary to provide a region corresponding to the region I _B. In addition, the area IV _A is photographed as the bed area in the distance image, but is actually outside the bed area. Therefore, the correction area as shown in FIGS. 11A and 11B is provided.

11A is an explanatory diagram illustrating an example of an image of a bed area including a correction area, and FIG. 11B is an explanatory diagram illustrating another example of an image of a bed area including a correction area, which is illustrated in FIGS. 4F and 4M, respectively. The distance image of the said bed area | region which added the said correction area | region to the bed upper surface area | region is shown.
When the bed area is set as shown in FIG. 11A and FIG. 11B, even if the subject person as shown in FIG. The risk of falling can be determined and notified more reliably.

  Next, the flow of calibrating and operating the danger notification device of the present invention will be described with reference to FIGS. 1A and 1B according to the step represented by S in the flowchart of FIG. 12A.

  In S <b> 101, the third party who monitors the subject can extract the bed 30 and the periphery of the bed 30 from above the bed 30 so that the upper surface of the bed 30 is extracted most widely in the distance image. The distance image acquisition means 10 is installed on the tripod 20 disposed in the position.

  In S102, the danger notification device 100 in which the distance image acquisition means 10 is installed by the third party is calibrated by the danger notification device calibration method of the present invention. Details of the calibration will be described later with reference to FIG. 12B.

  In S <b> 103, the danger notification device 100 operates to prevent the subject person from falling from the bed 30, and the process ends.

  Next, the flow of the calibration method of the danger notification device of the present invention will be described with reference to FIGS. 1A and 1B according to the step represented by S in the flowchart of FIG. 12B.

  In S201, the region setting unit 11 converts the distance image acquired by the distance image acquisition unit 101 from the distance image coordinate system to the camera coordinate system.

  In S202, the area setting unit 11 extracts the bed upper surface area based on the distance image.

  In S203, the area setting unit 11 calculates the camera posture in the world coordinate system, and the process proceeds to S204.

  In S204, the area setting unit 11 calculates the camera position in the world coordinate system, and the process proceeds to S205.

  In S205, the area setting unit 11 sets each area, and ends this process.

  Next, a flow for operating the danger notification device of the present invention will be described with reference to FIG. 1A and FIG. 1B according to the step represented by S in the flowchart of FIG. 12C.

  In S301, when the determination unit 12 determines that the number of pixels of the subject in the bedside region is smaller than a preset threshold value, the determination unit 12 continues to monitor the number of pixels of the subject in the bedside region. If the determination unit 12 determines that the number of pixels of the subject in the bedside region is equal to or greater than the threshold value, the process proceeds to S302. If the number of pixels of the target person in the bedside region is equal to or greater than a threshold value, any of the target person's fall, the target person's getting out of bed, and intrusion of the third party (visitor) other than the target person It may be.

  In S302, the determination unit 12 that has determined that the number of pixels of the subject in the bedside area is greater than or equal to a threshold value, determines that the number of pixels of the subject in the upper area is greater than the threshold value, To S301. If the determination unit 12 determines that the number of pixels of the subject in the upper region is equal to or less than the threshold value, the process proceeds to S303.

  In S303, the determination unit 12 that determines that the number of pixels of the subject in the upper region is equal to or less than the threshold value determines that the risk of the subject falling from the bed is high, and the notification unit 13 Operate it and notify a third party to finish this process.

  Examples of the present invention will be described below, but the present invention is not limited to these examples.

Example 1
<Calibration of danger alarm device>
<< Installation of distance image acquisition means >>
As shown in FIG. 1A, a tripod with a distance image sensor (Xtion PRO, manufactured by ASUS) as a distance image acquisition means 10 is installed on the long axis of the bed, and the distance image sensor is about 240 cm above the floor surface. The orientation was adjusted so that the upper surface of the rectangular bed was extracted most widely. Thereafter, a personal computer having a storage unit in which a program for operating the region setting unit, the determination unit, and the notification unit is stored is activated, and the distance image acquisition unit 10 performs a distance image of 320 pixels × 240 pixels as illustrated in FIG. Acquired.

<< Setting of each area >>
[Bed upper surface area extraction processing]
The distance image coordinates (u, v) in the acquired distance image were converted into camera coordinates _{Pu, v} by the following equation (1). In the following formula (1), the distance value k in the Z ₁ axis direction per luminance in the distance image is 12.5 mm / luminance, and the optical axis center coordinates (C _u , C _v ) are (160, 120), The ratio a between the focal length and the pixel size was 0.003452. I _{u, v} represents the luminance of each pixel corresponding to the distance image coordinates (u, v) in the imaging element of the distance image acquisition means.

-Generation of gradient images-
A set of camera coordinates P _{u, v} is created by removing the missing area from a small section (11 pixels × 11 pixels) centered on the distance image coordinates (u, v), and a unit of the third principal component of the created set The vector N _{u, v} was determined. However, since N _{u, v} exists in the point object, N _{u, v} is closer to the camera coordinates P _{u, v} (the one in which the inner product of P _{u, v} and N _{u, v} is positive). did.
By replacing the three components in the X ₁ axis direction, the Y ₁ axis direction, and the Z ₁ axis direction in the camera coordinate system of the obtained unit vector N _{u, v} with each element of RGB as shown in the following formula (2), A gradient image color-coded according to the direction of the unit vector _{Nu, v} as shown in FIG. 4A was generated.

-Color unevenness reduction correction for gradient images-
Since color unevenness occurred in the obtained gradient image, the obtained unit vector _{Nu, v} of each pixel was clustered by the k-means method with a division number T3 of 0.37 to correct the color unevenness of the gradient image. An image as shown in FIG. 4B was obtained.

-Extraction of bed upper surface area and floor area-
Among the classes acquired by the k-means method, pixels having the maximum area class were extracted, and an image obtained by extracting the bed upper surface area and the floor area as shown in FIG. 4C was obtained.

-Edge extraction-
In order to ensure separation of the bed upper surface area and the floor area in the image obtained by extracting the bed upper surface area and the floor area, a canny edge detection algorithm is applied to the distance image of FIG. An edge image in which the edge as shown in Fig. 1 was black and the other edges were white was obtained.

-Generation of composite image-
A composite image as shown in FIG. 4E is generated by multiplying the image obtained by extracting the bed upper surface area and floor area distance and the brightness value of each corresponding pixel of the edge image and multiplying them.

-Extraction of bed upper surface area-
Next, a labeling process for giving the same identification number to each pixel in the same area was performed on the composite image, and an area having the maximum number of pixels was extracted as the bed upper surface area, and an image as shown in FIG. 4F was obtained.

[Upper area setting process]
-Calculation of camera posture-
The distance value I _{u, v} of the distance image with respect to the pixel (u, v) on the bed upper surface area is acquired again _, and converted into the camera coordinates P _{u, v} by equation (1), and a small section (11 pixels × 11 pixels) ), The principal component analysis was performed again, and the unit vector _{Nu, v} of the third principal component of the principal component analysis was obtained. However, you select the _{N u, v} and _{P u, v} inner product of those who made a positive _{N u, v} here. Then, N _u, the unit square sum of distances from _v is the smallest vector N _u, a _v, was calculated as a camera tilt N.

-Calculation of camera position-
As shown in FIG. 7, the vector from the distance image acquiring unit 10 to any point of the upper surface of the bed 30 when the P _i, distance world coordinate system from the image acquisition unit 10 to any point of the upper surface of the bed 30 The height l _i was calculated by the following formula (3), and the average value l of the height l _i was calculated by the following formula (5). As a result, l = 1,718 mm.
Also, when the vector to an arbitrary point of the floor surface and Q _i from the distance image acquiring unit, the height m _i in the world coordinate system from the distance image acquiring unit 10 to the floor 40 by the following equation (4) calculated, to calculate the mode m height _{m i} by the following equation (6) results were m = 2,460mm.

In the above formula (5), Average () represents an average value.

However, Mode () in the above formula (6) represents the mode value.

-Setting the upper area-
In the distance image, after calculating the camera coordinates P _{u, v according} to the above equation (1) _, a luminance difference g ₁ between the upper surface of the bed and the floor surface is calculated as 60 according to the following equation (7). the brightness g ₂ surface as 250 seeking I1 _{u, v,} to give the Z ₀ range image and brightness corresponding to the distance image to the Z ₀ axis component in the world coordinate system.

[Correction area settings]

(I) Among the RGB components of the distance image coordinates (u, v) in the correction area, the luminance value of the B component is represented by Ib _{u, v} , and the luminance value of the R component is represented by Ir _{u, v} . If the distance image coordinates (u, v) of the bed upper surface area, _{Ir u, v} and _{Ib u,} a _v, is initialized as 255 and _g 2 -g ₁ respectively. Otherwise, Ir _{u, v} and Ib _{u, v} are initialized as 0.

(Ii) Select one pixel of coordinates (u, v) from within the bed area, obtain the luminance value I _{u, v} at the coordinates (u, v) of the distance image, and use the camera coordinates in the above equation (1) The values P _{u, v} were calculated.

(Iii) b the upper surface of the bed in virtually _{Z 0} axis direction was determined (initially and b = 1.) Range image coordinates when you move (u2, v2). The camera coordinate values P2 _{u, v} = (P2 _x , P2 _y , P2 _z ) after movement were calculated by the following formula (10) and the following formula (11). The following equation (11) was obtained by solving for u and v in the above equation (1).

(Iv) Further, the Z distance image value at the distance image coordinates (u2, v2) when moving virtually in the _Z0 axis direction is I2u _{, v,} and I2u _{, v} is derived from the following equation (12). did.

(V) The pixel value in the image of the correction area was updated by the following formula (13) and the following formula (14).

(Vi) b was counted and (iii) to (v) were repeated until b = 1,800 mm, which is about the height of a person.

(Vii) Another pixel in the bed upper surface area was selected, and (ii) to (vi) were performed on all the pixels in the bed upper surface area.

[Bedside area setting process]
Next, in order to determine whether a part of the subject's body exists in the bedside region or in the bedside lower region, with the distance threshold constant e being 300 mm, according to the following equation (14): When determination of the _{Z 0} axis direction brightness threshold _{I T2,} was 225.7. As described above, the danger alarm device was calibrated.

<Evaluation of danger alarm device>
Using the danger notification device 100 that has been calibrated as described above, the threshold T1 of the number of pixels of the subject in the bedside region is set to 2,000 pixels, and the number of pixels of the subject in the upper region is set. When the threshold value T2 is 1,000 pixels, the threshold value is equal to or greater than the threshold value T1, and the threshold value is equal to or less than the threshold value T2, it is determined that the subject has a high risk of falling from the bed, and a warning sound is generated. I made it.
The operation monitored by the danger notification device is a case where the target person performs a picking-up operation in the state of bed as shown in FIG. 13, and is a warning sound that notifies the target person that the risk of falling is high? I evaluated it. As a result of the evaluation, a warning sound was emitted to the subject, and the risk of the subject falling from the bed could be more reliably determined and notified.

(Example 2)
In Example 1, the operation monitored by the danger notification device is performed when the subject performs a pick-up operation in the bed state as shown in FIG. 13, and then the bed in the bed state as shown in FIG. Evaluation was performed in the same manner as in Example 1 except that the operation was changed when the fence was operated. As a result of the evaluation, a warning sound was emitted to the subject after the operation of the bed fence, and the risk of the subject falling from the bed could be more reliably determined and notified.

(Example 3)
Example 1 is different from Example 1 except that the target person as shown in FIG. 13 performs the picking-up operation in the bedded state to the case where the target person performs the leaving action as shown in FIG. 15. Evaluation was performed in the same manner. As a result of the evaluation, no warning sound was emitted to the subject, and the risk of the subject falling from the bed could be more reliably determined and notified.

Example 4
Example 1 is the same as Example 1 except that the target person as shown in FIG. 13 picked up in the bedside state and changed to the intrusion of a third party from the outside as shown in FIG. Evaluated. As a result of the evaluation, no warning sound was emitted to the subject, and the risk of the subject falling from the bed could be more reliably determined and notified.

(Example 5)
<Evaluation of calibration method for danger alarm device>
The result of having calibrated the position where the distance image can be obtained as shown in FIG. 2 by the calibration method of the danger notification device of the present invention after the distance image sensor is installed at the position where the distance image can be obtained as shown in FIG. 17A Is shown in FIGS. 17A and 17B. As shown in FIGS. 17A and 17B, the area was divided at the boundary between the bed upper surface and the bed rail, and the bed upper surface area could be extracted with high accuracy.

(Comparative Example 1)
In Example 5, the method for calibrating the danger notification device according to the present invention is changed from the conventional watershed method, a point marked with ○ is designated as a reference point, and a point where the luminance value of an adjacent pixel changes by a predetermined luminance value or more is designated. Evaluation was performed in the same manner as in Example 5 except that the boundary was determined and the area was divided and the bed upper surface area was extracted, and the results are shown in FIGS. 18A and 18B. As shown in FIG. 18A and FIG. 18B, it was not possible to divide at the boundary between the bed upper surface and the bed rail, and the bed upper surface area could not be extracted accurately.

(Example 6)
In the fifth embodiment, the position where the distance image can be acquired as shown in FIG. 17A is changed to the position where the distance image can be acquired as shown in FIG. 19A. The results are shown in FIGS. 19A and 19B. As shown in FIGS. 19A and 19B, the upper surface area of the bed could be extracted with high accuracy by dividing the boundary between the upper surface of the bed and the bed rail.

(Comparative Example 2)
In Example 6, the evaluation was performed in the same manner as in Example 6 except that the method for calibrating the danger notification device of the present invention was changed to the conventional watershed method, and the results are shown in FIGS. 20A and 20B. As shown in FIG. 20A and FIG. 20B, it was not possible to divide at the boundary between the bed upper surface and the bed rail, and the bed upper surface area could not be extracted accurately.

(Example 7)
In the fifth embodiment, the position where the distance image can be acquired as shown in FIG. 17A is changed to the position where the distance image can be acquired as shown in FIG. 21A. The results are shown in FIG. 21A and FIG. 21B. As shown in FIG. 21A and FIG. 21B, the upper surface area of the bed was extracted with high accuracy by dividing the boundary between the upper surface of the bed and the bed rail.

(Comparative Example 3)
In Example 7, evaluation was made in the same manner as in Example 7 except that the method for calibrating the danger notification device of the present invention was changed to the conventional watershed method, and the results are shown in FIGS. 22A and 22B. As shown in FIG. 22A and FIG. 22B, it was not possible to divide at the boundary between the bed upper surface and the bed rail, and the bed upper surface area could not be extracted with high accuracy.

DESCRIPTION OF SYMBOLS 10 Distance image acquisition means 11 Area | region setting means 12 Determination means 13 Notification means 14 Control means 15 Information processing apparatus 20 Tripod 30 Bed 40 Floor 50 Ceiling 100 Danger notification apparatus

Claims (14)

Distance image acquisition means for capturing the bed and the periphery of the bed from above the bed, and acquiring a distance image using the distance value to the object to be imaged as a pixel value;
An area setting means for setting, as a bed area, a bed upper surface area extracted from the acquired distance image, and a correction area for preventing erroneous detection of the long sitting state of the object person when the object person is on the bed; ,
Determining means for determining whether or not there is a risk that the subject falls from the bed based on the number of pixels of the subject outside the bed region of the distance image including the subject;
Informing means for informing the determination result of the determining means to at least one of the target person and a third party;
A danger notification device characterized by comprising: The area setting means is outside the bed area,
A bedside area for detecting protrusion of the subject from the bed;
An upper region of the bed region and the bedside region;
Further set
The determination means determines whether or not there is a risk that the subject falls from the bed based on at least one of the number of pixels of the subject in the bedside region and the number of pixels of the subject in the upper region. The danger notification device according to claim 1.   In the case where the number of pixels of the subject in the bedside region is equal to or greater than a predetermined number of pixels, the subject is determined when the number of pixels of the subject in the upper region is equal to or less than a predetermined number of pixels. The danger notification device according to claim 2, wherein it is determined that there is a risk of falling from the bed. The region setting means further sets the region above the bed upper surface region in the upper region as the correction region;
The danger notification device according to any one of claims 2 to 3, wherein the determination unit determines not to erroneously detect the bed leaving state of the subject and the third party. A distance image obtaining step of photographing the bed and the periphery of the bed from above the bed, and obtaining a distance image with a distance value to the photographing object as a pixel value;
An area setting step for setting, as a bed area, a bed upper surface area extracted from the acquired distance image and a correction area for preventing erroneous detection of the long sitting state of the object person when the object person is on the bed; ,
A determination step of determining whether or not there is a risk that the subject falls from the bed based on the number of pixels of the subject outside the bed region of the distance image including the subject;
A notification step of notifying the determination result of the determination step to at least one of the target person and a third party;
A danger notification method comprising: Distance image acquisition means for capturing the bed and the periphery of the bed from above the bed, and acquiring a distance image using the distance value to the object to be imaged as a pixel value;
Area setting means for setting the bed upper surface area extracted from the acquired distance image;
A calibration method for a danger notification device having
A risk notification including a bed upper surface area extracting step of generating a gradient image obtained by dividing the acquired distance image into regions having similar spatial gradients and extracting the bed upper surface area based on the gradient image. Device calibration method.   The bed upper surface region extraction step further generates an edge image representing a portion where the amount of change in the pixel value of the distance image is large, and the bed upper surface area based on a composite image of the edge image and the gradient image The method for calibrating a danger notification device according to claim 6, comprising a process of extracting a region. In the distance image acquired by the distance image acquisition means, further includes an installation step of installing the distance image acquisition means so that a region corresponding to the upper surface of the bed is extracted in a rank of a predetermined number of pixels,
The method for calibrating a danger notification device according to claim 7, wherein the bed upper surface area extraction step includes a process of extracting an area having the predetermined number of pixels as the bed upper surface area.   The method for calibrating a danger notification device according to any one of claims 7 to 8, further comprising a calculation step of calculating the position and orientation of the distance image acquisition means based on the bed upper surface area.   The danger notification device calibration method according to claim 9, further comprising a region setting step of setting a region by dividing a space in the camera coordinate system based on the bed upper surface region.   The method of calibrating a danger notification device according to claim 10, wherein the region setting step sets a correction region for preventing erroneous detection of the subject's long sitting position when the subject is on the bed.   The danger notification according to claim 11, wherein the region setting step further sets a region above the bed upper surface region in the upper region as a correction region for preventing erroneous detection of the person's bed leaving state and a third party. Device calibration method. Distance image acquisition means for capturing the bed and the periphery of the bed from above the bed, and acquiring a distance image using the distance value to the object to be imaged as a pixel value;
Area setting means for setting the bed upper surface area extracted from the acquired distance image;
A calibration method for a danger notification device having
Based on the upper surface area of the bed, the space in the camera coordinate system is divided to set the area, and when the target person is on the bed among the set areas, the long sitting state of the target person is not erroneously detected. A method for calibrating a danger notification device, comprising an area setting step for setting a correction area for the purpose.   The method for calibrating a danger notification device according to claim 13, further comprising a calculation step of calculating a position and orientation of the distance image acquisition unit based on the bed upper surface area.