Gas migration estimation device, gas migration estimation method, estimation model generation device, estimation model generation method, and program
An estimation device and an estimation method are capable of detecting whether or not gas is flowing in a distal-proximal direction from a viewpoint even in a case where gas monitoring is performed on the basis of only an image from one viewpoint. The gas migration estimation device includes: an image input unit that receives, as an input, a gas distribution moving image in which a presence range of gas leaked into a space is indicated as a gas region; and a gas migration estimation unit that estimates a migration state of gas in a distal-proximal direction corresponding to the gas distribution moving image received by the image acquisition unit, by using an estimation model that is machine-learned using a combination of a training gas distribution moving image and a migration state of gas in a distal-proximal direction in the training gas distribution moving image as training data.
1 . A gas migration estimation device comprising:
an image inputter that receives, as an input, a gas distribution moving image including a plurality of frames and in which a presence range of gas leaked into a space is indicated as a gas region; and
a gas migration estimator that estimates a migration state of gas in a distal-proximal direction corresponding to the gas distribution moving image received by a hardware processor, by using an estimation model that is machine-learned using a combination of a training gas distribution moving image and a migration state of gas in a distal-proximal direction in the training gas distribution moving image as training data, to determine whether or not the gas is flowing in the distal-proximal direction, which is a depth direction of the gas distribution moving image.
2 . The gas migration estimation device according to claim 1 , wherein the gas migration estimator uses a relative distal-proximal velocity indicating a migration velocity of the gas in the distal-proximal direction as a relative value with respect to a migration velocity in a viewing angle direction, as the migration state of the gas in the distal-proximal direction in the training data.
3 . The gas migration estimation device according to claim 1 , wherein the gas migration estimator outputs a migration velocity of the gas in the distal-proximal direction in the gas distribution moving image received by the image inputter as the migration state of the gas in the distal-proximal direction.
4 . The gas migration estimation device according to claim 1 , wherein the gas migration estimator outputs a probability distribution of a migration velocity of the gas in the distal-proximal direction in the gas distribution moving image received by the image inputter as the migration state of the gas in the distal-proximal direction.
5 . The gas migration estimation device according to claim 1 , wherein the image inputter includes an imaging means that senses infrared light, and the gas region is an image of gas that absorbs the infrared light, or water vapor.
6 . The gas migration estimation device according to claim 1 , wherein the image inputter includes an imaging means that senses visible light, and the gas region is an image of gas that absorbs the visible light, or water vapor.
7 . The gas migration estimation device according to claim 1 , wherein the image inputter receives, as an input, a gas distribution moving image obtained by extracting a specific frequency component from a moving image obtained by imaging a space.
8 . The gas migration estimation device according to claim 1 , further comprising an image display that superimposes and displays the migration state of the gas in the distal-proximal direction on the gas distribution moving image received by the image inputter.
9 . The gas migration estimation device according to claim 8 , wherein the image display displays at least one of presence or absence of migration of the gas in the distal-proximal direction, a migration direction, and a migration velocity by color.
10 . The gas migration estimation device according to claim 8 , wherein the image display displays at least one of presence or absence of migration of the gas in the distal-proximal direction, a migration direction, and a migration velocity by a length of a line or an arrow.
11 . The gas migration estimation device according to claim 8 , further comprising a three-dimensional flow estimator that estimates a migration state of gas in a viewing angle direction in the gas distribution moving image received by the image inputter, and estimates a migration state of gas as three-dimensional information from the migration state of the gas in the viewing angle direction and the migration state of the gas in the distal-proximal direction,
wherein the image display displays the migration state of the gas as the three-dimensional information.
12 . An estimation model generation device comprising:
an image inputter that receives, as an input, a gas distribution moving image including a plurality of frames and in which a presence range of gas leaked into a space is indicated as a gas region;
a velocity inputter that receives, as an input, a migration state of gas in a distal-proximal direction in the gas distribution moving image; and
a hardware processor that performs machine learning using a combination of the gas distribution moving image and the migration state of the gas in the distal-proximal direction as training data, and generates an estimation model that outputs the migration state of the gas in the distal-proximal direction using the gas distribution moving image as an input, to indicate whether or not the gas is flowing in the distal-proximal direction, which is a depth direction of the gas distribution moving image.
13 . A gas migration estimation device comprising:
an image inputter that receives, as an input, a gas distribution moving image including a plurality of frames and in which a presence range of gas leaked into a space is indicated as a gas region;
a second hardware processor that performs machine learning using a combination of a training gas distribution moving image and a migration state of gas in a distal-proximal direction in the training gas distribution moving image as training data and that generates an estimation model; and
a gas migration estimator that estimates a migration state of gas in a distal-proximal direction corresponding to the gas distribution moving image received by a first hardware processor, by using the estimation model, to determine whether or not the gas is flowing in the distal-proximal direction, which is a depth direction of the gas distribution moving image.
14 . A gas migration estimation method comprising:
receiving, as an input, a gas distribution moving image including a plurality of frames and in which a presence range of gas leaked into a space is indicated as a gas region; and
estimating a migration state of gas in a distal-proximal direction corresponding to the gas distribution moving image received by using an estimation model that is machine-learned using a combination of a training gas distribution moving image and a migration state of gas in a distal-proximal direction in the training gas distribution moving image as training data, to determine whether or not the gas is flowing in the distal-proximal direction, which is a depth direction of the gas distribution moving image.
15 . A non-transitory recording medium storing a computer readable program causing a computer to perform gas migration estimation processing, the gas migration estimation processing comprising:
receiving, as an input, a gas distribution moving image including a plurality of frames and in which a presence range of gas leaked into a space is indicated as a gas region; and
estimating a migration state of gas in a distal-proximal direction corresponding to the gas distribution moving image received by using an estimation model that is machine-learned using a combination of a training gas distribution moving image and a migration state of gas in a distal-proximal direction in the training gas distribution moving image as training data, to determine whether or not the gas is flowing in the distal-proximal direction, which is a depth direction of the gas distribution moving image.
16 . An estimation model generation method comprising:
receiving, as an input, a gas distribution moving image including a plurality of frames and in which a presence range of gas leaked into a space is indicated as a gas region;
receiving, as an input, a migration state of gas in a distal-proximal direction in the gas distribution moving image; and
performing machine learning using a combination of the gas distribution moving image and the migration state of the gas in the distal-proximal direction as training data, and generating an estimation model that outputs the migration state of the gas in the distal-proximal direction using the gas distribution moving image as an input, to indicate whether or not the gas is flowing in the distal-proximal direction, which is a depth direction of the gas distribution moving image.
17 . A non-transitory recording medium storing a computer readable program causing a computer to perform estimation model generation processing, the estimation model generation processing comprising:
receiving, as an input, a gas distribution moving image including a plurality of frames and in which a presence range of gas leaked into a space is indicated as a gas region;
receiving, as an input, a migration state of gas in a distal-proximal direction in the gas distribution moving image; and
performing machine learning using a combination of the gas distribution moving image and the migration state of the gas in the distal-proximal direction as training data, and generating an estimation model that outputs the migration state of the gas in the distal-proximal direction using the gas distribution moving image as an input, to indicate whether or not the gas is flowing in the distal-proximal direction, which is a depth direction of the gas distribution moving image.
18 . A gas migration estimation method comprising:
receiving, as an input, a gas distribution moving image including a plurality of frames and in which a presence range of gas leaked into a space is indicated as a gas region;
generating an estimation model by performing machine learning using a combination of a training gas distribution moving image and a migration state of gas in a distal-proximal direction in the training gas distribution moving image as training data; and
estimating a migration state of gas in a distal-proximal direction corresponding to the gas distribution moving image received by using the estimation model, to determine whether or not the gas is flowing in the distal-proximal direction, which is a depth direction of the gas distribution moving image.
19 . A non-transitory recording medium storing a computer readable program causing a computer to perform gas migration estimation processing, the gas migration estimation processing comprising:
receiving, as an input, a gas distribution moving image including a plurality of frames and in which a presence range of gas leaked into a space is indicated as a gas region;
generating an estimation model by performing machine learning using a combination of a training gas distribution moving image and a migration state of gas in a distal-proximal direction in the training gas distribution moving image as training data; and
estimating a migration state of gas in a distal-proximal direction corresponding to the gas distribution moving image received by using the estimation model, to determine whether or not the gas is flowing in the distal-proximal direction, which is a depth direction of the gas distribution moving image.
20 . The gas migration estimation device according to claim 2 , wherein the gas migration estimator outputs a migration velocity of the gas in the distal-proximal direction in the gas distribution moving image received by the image inputter as the migration state of the gas in the distal-proximal direction.
21 . The gas migration estimation device according to claim 1 , wherein the image inputter receives, as the input, the gas distribution moving image including the plurality of frames, which are all obtained from a single viewpoint.