JP7438470B2 - Person detection device, person detection method, and person detection program - Google Patents

Description

The present disclosure relates to person detection technology.

Patent Document 1 discloses an image monitoring device that sequentially acquires thermal images of a monitoring space, performs image processing on the thermal images, and determines the presence or absence of a target moving object in the monitoring space, and detects environmental changes from the thermal images. a background temperature change determination unit that determines an image area caused by a temperature change of a background object as a background temperature change area; an extraction unit that extracts a temperature change area from the thermal image so that the background temperature change area is not included; a determination unit that determines the presence or absence of the target moving object based on the characteristics of the temperature change area, and the background temperature change determination unit includes A technology related to an image monitoring device characterized in that, in a change factor determination period that is a period, an image area showing a temperature change of only a temperature increase or an image area showing a temperature change of only a temperature drop is determined as the background temperature change area. Disclosed.

Japanese Patent Application Publication No. 2013-207392

According to the technique disclosed in Patent Document 1, a region removed from a thermal image generated as a background temperature change region is a region exhibiting only a temperature increase or decrease over a predetermined period due to an environmental change. Here, the temperature change due to environmental change is generally a gradual temperature change. On the other hand, in electronic devices that rapidly generate heat, such as monitors or multifunction devices, or electrical devices, such as electric heaters, temperature changes occur instantaneously, unlike temperature changes due to environmental changes. Therefore, according to the technology of Patent Document 1, it is difficult to determine these electronic devices or electrical devices as background temperature change areas and remove them from the thermal image, and there is a possibility that they may be mistakenly detected as people. There is a problem.

The present disclosure has been made to solve such problems, and aims to provide a person detection technique that can detect a person while suppressing false detection of an imaged object.

One aspect of the person detection device according to the embodiment of the present disclosure is a time series region that extracts a time series of partial regions of a plurality of thermal images or heat maps as a time series region from a plurality of thermal images or heat maps at different times. an extraction unit, a variation index value calculation unit that calculates the value of a variation index that quantifies the temperature distribution or the label distribution for each partial region included in the extracted time series region, and the calculated variation index value. a fluctuation characteristic calculation unit that calculates a value of an irregularity index indicating the irregularity of the fluctuation and a value of a scale index indicating the magnitude of the fluctuation; person determination that determines that a person is included in the extracted time series region when the value of the calculated scale index is equal to or higher than a predetermined scale index threshold; It is equipped with a section and a section.

According to the person detection device according to the embodiment of the present disclosure, a person can be detected while suppressing false detection of an imaged object.

1 is a diagram showing functional blocks of a person detection device according to Embodiment 1. FIG. FIG. 3 is a schematic diagram showing thermal images at predetermined intervals. FIG. 2 is a diagram showing an example of a hardware configuration of a person detection device. FIG. 2 is a diagram showing an example of a hardware configuration of a person detection device. It is a flowchart of a person detection method.

Hereinafter, various embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. Note that components given the same or similar symbols in the drawings have the same or similar configurations or functions, and overlapping explanations of such components will be omitted.

Embodiment 1.
<Configuration>
A person detection device according to Embodiment 1 of the present disclosure will be described with reference to FIGS. 1 and 2. As shown in FIG. 1, the person detection device 10 is connected to an imaging device 20 that images an imaging space. The imaging device 20 is, for example, a thermal camera that is arranged to image a room and generates a thermal image of the temperature distribution in the imaging space. The person detection device 10 acquires a thermal image generated by the imaging device 20 from the imaging device 20, and detects whether the object imaged in the acquired thermal image is a person.

Instead of a thermal image, the person detection device 10 may obtain a heat map in which the degree of personhood is evaluated and labeled for each pixel. In this case, the imaging device 20 may be an RGB camera, a heat map creation device (not shown) creates a heat map from the image generated by the RGB camera using semantic segmentation technology, and the person detection device 10 uses the heat map is obtained from the heat map creation device. Although a heat map is a map that has already evaluated the degree of personhood, the technology of the present disclosure can be used to further evaluate whether an object that has been evaluated as a person by the heat map creation device is really a person. . The following will explain the case where the image acquired by the person detection device 10 is a thermal image.

In order to detect whether the object to be imaged is a person, the person detection device 10 includes, for example, a time series region extraction section 11, a variation index value calculation section 12, a variation characteristic calculation section 13, a learning result utilization section 14, and a person determination section 15. Note that the person detection device 10 includes a buffer memory (not shown), the thermal image generated by the imaging device 20 is temporarily stored in the buffer memory, and the time-series area extraction unit 11 acquires the thermal image from the buffer memory.

(Time series region extraction part)
The time-series region extraction unit 11 extracts a time-series partial region of a thermal image as a time-series region from a thermal image stored in a buffer memory (not shown). A partial area is a specific area included in a thermal image, and is narrower than the entire area of the thermal image. The time series region extraction unit 11 generates a time series of partial regions by extracting the same partial region from a plurality of thermal images.

The time-series region extraction unit 11 acquires thermal images at predetermined intervals, such as intervals of several tens of seconds, one minute, or several minutes, from the buffer memory. The thermal images stored in the buffer memory themselves may be thermal images generated at predetermined intervals, or the intervals at which thermal images are extracted from the thermal images continuously stored in the buffer memory may be at predetermined intervals. It may be.

FIG. 2 is a schematic diagram showing thermal images at predetermined intervals. A thermal image D-t1 at time t1, a thermal image D-t2 at time t2, a thermal image D-t3 at time t3, and a thermal image D-t4 at time t4 are shown. Each interval from time t1 to t4 is a predetermined interval. The thermal image shows a seated person. The temperature of each pixel of the thermal image is expressed in gray scale, with the highest temperature as white and the lowest temperature as black. Partial regions are indicated by square boxes. The center of gravity of the temperature value of the partial region is indicated by an "x" mark.

In one aspect, the partial area may be a neighboring area consisting of a pixel at the specified coordinates and pixels in its vicinity (see FIG. 2). The number of neighboring pixels may be 8 or 24. The coordinates may be specified by external input by the user of the person detection device 10, or the time-series region extracting unit 11 may identify both a high-temperature region having a temperature equal to or higher than a predetermined high-temperature threshold and a low-temperature region less than the high-temperature threshold. It is also possible to search for an area containing the area and designate the area to be included. As the threshold value, for example, a temperature corresponding to the lower limit of body temperature, for example 35.0° C., may be set as the high temperature threshold value. Further, as another example, since the heat radiation of a person who is far away becomes weaker, a temperature lower than 35.0° C. may be set as the high temperature threshold.

In another aspect, a plurality of partial regions may be extracted from one thermal image. The plurality of partial regions may be obtained discretely from one thermal image, or may be continuously obtained from one thermal image by raster scanning. The size of the partial area may be the same size as the aforementioned neighboring area.

In another aspect, the partial region may include both a part of the high temperature region having a temperature equal to or higher than a predetermined high temperature threshold and a low temperature region less than the high temperature threshold. In other words, the partial region may include a part of the boundary between the high temperature region and the low temperature region. In this way, by focusing on only a part of the boundary between the high temperature region and the low temperature region, it is possible to better observe the movement of the person who is the detection target.

If a partial region includes only a high-temperature region at both a certain time and another time, even if the object moves between those times, it is difficult to detect large fluctuations in the high-temperature region, so the movement of the object cannot be detected. It is difficult to determine that the person is a person. On the other hand, if you observe a region that includes both a part of the high-temperature region and a low-temperature region at any given time, that is, if you set the partial region to also include the background, which is the low-temperature region, you can see the object in contrast with the background. This makes it easier to capture the movement of the object. In other words, it becomes easier to grasp fluctuations in the values of fluctuation indicators such as temperature center of gravity, which will be described later.

In addition, even if a partial region includes all of the high temperature region and a low temperature region at both one time and another time, the object is moving between those times, as in the case where the partial region includes only the high temperature region. However, it is difficult to grasp the fluctuations.

Moreover, the amount of calculation can be suppressed by acquiring time-series regions only for neighboring regions centered on one or more designated pixels, instead of performing raster scanning to acquire time-series regions.

The time series region extraction section 11 outputs the extracted time series region to the variation index value calculation section 12 and the learning result utilization section 14.

(Fluctuation index value calculation unit)
The variation index value calculation unit 12 calculates the value of a variation index that quantifies the temperature distribution for each partial region included in the extracted time series region. Examples of variation indicators include the coordinates of the temperature centroid. Note that in the case of a heat map, examples of the variation index include the centroid of the label value labeled on the pixel of the heat map.

The value of the coordinate of the temperature center of gravity is obtained by weighting the value of each coordinate defining the partial area by the weight of the heat in the partial area that the coordinate has. For example, if the temperature center of gravity x _G on the X axis in the XY orthogonal coordinates is determined by x _G =Σx _n m _n /Σm _n . Here, n is a natural number determined by the width of the partial region, x _n is the coordinate of the X axis, and m _n is the sum of temperatures at the coordinate x _n . The temperature center of gravity _yG can be found in the same manner for the Y-axis.

As a modified example, the area exhibiting a temperature equal to or higher than a predetermined high temperature threshold is set as 1, and the other areas as 0, and the value of the coordinates of the temperature center of gravity is calculated using the ratio of 1 in the partial area or the values of 1 and 0. may be calculated. Note that in the case of a heat map, processing may be performed in the same manner as in the case of a thermal image, with areas to which a person label representing a person is attached set to 1, and other areas to 0.

(Variation characteristic calculation section)
The fluctuation characteristic calculation unit 13 calculates the value of an index for capturing the characteristics of fluctuation of the object included in the time-series region. Specifically, the fluctuation characteristic calculation unit 13 calculates the value of an index for determining whether the fluctuation of the object is irregular and large-scale. More specifically, the fluctuation characteristic calculation unit 13 calculates the value of the irregularity index indicating the irregularity of the fluctuation and the magnitude of the fluctuation for the value of the fluctuation index calculated by the fluctuation index value calculation unit 12. Calculate the value of the scale index shown.

As an example of a technique for determining irregularity, the p-value of a trend test such as the Cox-Stuart test may be calculated. The Cox-Start test is a method for testing the increasing or decreasing tendency of multiple measured values (see Shinsuke Muto, "Statistical Analysis Handbook", Asakura Shoten, 1995). Furthermore, the p-value is a threshold value of the probability that it can be judged as random, and the larger the p-value, the more irregular it is. When the variation index is the temperature center of gravity, the variation characteristic calculation unit 13 may calculate the p value for each of the X coordinate and Y coordinate of the temperature center of gravity.

As an example of a method for determining large scale, a time-series variation index may be subjected to linear regression, and the standard deviation of the difference between the regressed value and the variation index may be used. When the fluctuation index is the temperature center of gravity, the fluctuation characteristic calculation unit 13 performs linear regression on each of the X and Y coordinates of the temperature center of gravity to calculate the standard deviation value of the X coordinate of the temperature center of gravity and the standard deviation of the Y coordinate of the temperature center of gravity. Calculate the deviation value.

(Learning Results Utilization Department)
The learning result utilization unit 14 is a functional unit that uses the learned model to determine whether a person is shown in the time series area supplied from the time series area extraction unit 11. The trained model is a trained model that has learned whether the object shown in the time-series area is a person by machine learning or deep learning, and is held in a memory (not shown). The learning result utilization unit 14 inputs the time series region supplied from the time series region extraction unit 11 to a trained model held in a memory (not shown), and extracts the object appearing in the input time series region. A determination result indicating the degree of personhood of is obtained from the trained model. The learning result utilization unit 14 supplies the determination result indicating the degree of personhood to the person determination unit 15.

(Person Determination Department)
The person determining unit 15 is a functional unit that determines whether the object shown in the time-series area is a person. As an example, the person determination unit 15 may determine whether the calculated irregularity index value is equal to or greater than a predetermined irregularity index threshold, and the calculated scale index value is equal to or greater than a predetermined scale index threshold. In certain cases, it is determined that a person is included in the extracted time series region. The person determination unit 15 may perform determination using a threshold value for both the X coordinate and the Y coordinate, or may perform determination for only one of these coordinates. Further, the person determining unit 15 may determine that the high temperature region in the extracted time series region is a person.

As another example, the person determination unit 15 may perform person determination on the extracted time-series region based also on the determination result by the learning result utilization unit 14. That is, the person determination unit 15 determines that the value of the calculated irregularity index is greater than or equal to the irregularity index threshold, the value of the calculated scale index is greater than or equal to the scale index threshold, and the learning result utilization unit 14 It may be determined that the object included in the extracted time-series region is a person when the person degree indicated by the determination result is equal to or higher than a predetermined person degree threshold.

As another example, the person determination unit 15 may perform person determination using a decision tree. For example, the calculated irregularity index value is within a predetermined irregularity index value range, the calculated scale index value is within a predetermined scale index value range, and If the person degree indicated by the determination result by the learning result utilization unit 14 is within a predetermined person degree range, it may be determined that the object included in the extracted time-series region is a person. In this way, by setting detailed determination conditions based on each input value to the person determination unit 15, the accuracy of determination can be improved.

When determining that a person is included in the extracted time-series region, the person determination unit 15 outputs an identification result indicating that the object in the extracted time-series region is a person. The output identification result is used for control by a device such as an air conditioner (not shown).

Next, an example of the hardware configuration of the person detection device 10 will be described with reference to FIGS. 3A and 3B. The functions of the time-series region extraction section 11, the variation index value calculation section 12, the variation characteristic calculation section 13, the learning result utilization section 14, and the person determination section 15 are realized by processing circuitry. The processing circuitry, even if it is a dedicated processing circuit 100a as shown in FIG. 3A, is stored in a memory 100c, which is a storage medium, as shown in FIG. 3B. It may also be a processor 100b that executes a program.

When the processing circuitry is a dedicated processing circuit 100a, the dedicated processing circuit 100a may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, or an application specific integrated circuit (ASIC). , FPGA (field-programmable gate array), or a combination thereof. The functions of each part of the time series region extraction section 11, variation index value calculation section 12, variation characteristic calculation section 13, learning result utilization section 14, and person determination section 15 are realized by a plurality of separate processing circuits. Alternatively, the functions of each part may be realized by a single processing circuit.

When the processing circuitry is the processor 100b, each function of the time series region extraction section 11, variation index value calculation section 12, variation characteristic calculation section 13, learning result utilization section 14, and person determination section 15 is implemented by software, Realized by firmware or a combination of software and firmware. That is, each function of the variation index value calculation section 12, the variation characteristic calculation section 13, the learning result utilization section 14, and the person determination section 15 is realized by a computer including a processor 100b and a memory 100c. Software and firmware are written as programs and stored in memory 100c. The processor 100b realizes the functions of each section by reading and executing programs stored in the memory 100c. Here, examples of the memory 100c include RAM (random access memory), ROM (read-only memory), flash memory, EPROM (erasable programmable read only memory), EEPROM (electrically erasable programmable read-only memory), etc. These include nonvolatile or volatile semiconductor memory, magnetic disks, flexible disks, optical disks, compact disks, minidisks, and DVDs.

Note that some of the functions of the time-series region extraction unit 11, variation index value calculation unit 12, variation characteristic calculation unit 13, learning result utilization unit 14, and person determination unit 15 are realized by dedicated hardware, and integrated. The portion may be realized by software or firmware. In this way, the processing circuit can implement each of the above functions using hardware, software, firmware, or a combination thereof.

<Operation>
Next, the operation of the person detection device 10 will be described with reference to FIG. 4.

In step ST1, the time-series region extracting unit 11 extracts a time-series partial region of a plurality of thermal images as a time-series region from a plurality of thermal images at different times.

In step ST2, the variation index value calculation unit 12 calculates the value of a variation index that quantifies the temperature distribution or the label distribution for each partial region included in the time series region extracted by the time series region extraction unit 11. do.

In step ST3, the variation characteristic calculation unit 13 calculates the value of the irregularity index indicating irregularity and the value of the scale index indicating the size of the calculated variation index value.

In step ST5, the person determination unit 15 determines that the value of the calculated irregularity index is greater than or equal to a predetermined irregularity index threshold, and the value of the calculated scale index is greater than or equal to a predetermined scale index threshold. If , it is determined that a person is included in the extracted time series region.

When the person detection device 10 includes the learning result utilization unit 14, the person detection device 10 performs the process of step ST4 in parallel to the processes of steps ST2 and ST3. In step ST4, the learning result utilization unit 14 inputs the time series region extracted by the time series region extraction unit 11 into the learned model, and calculates the degree of personhood of the object included in the time series region from the learned model. Get the results shown. The learning result utilization unit 14 supplies the acquired results to the person determination unit 15. In this case, in step ST5, the person determination unit 15 determines that the value of the calculated irregularity index is equal to or higher than the irregularity index threshold, the value of the calculated scale index is equal to or higher than the scale index threshold, and When the degree of personhood indicated by the determination result by the learning result utilization unit 14 is equal to or greater than a predetermined personhood degree threshold, it is determined that the object included in the extracted time-series region is a person.

<Additional notes>
Some aspects of the various embodiments described above are summarized as follows.

(Additional note 1)
The person detection device (10) of Supplementary note 1 includes a time-series region extraction unit ( 11), a fluctuation index value calculation unit (12) that calculates the value of a fluctuation index that quantifies the temperature distribution or label distribution for each partial region included in the extracted time series region, and the calculated fluctuation A fluctuation characteristic calculation unit (13) that calculates the value of an irregularity index indicating the irregularity of fluctuation and the value of a scale index representing the magnitude of the fluctuation regarding the value of the index, and the calculated irregularity index. If the value of is greater than or equal to a predetermined irregularity index threshold, and the value of the calculated scale index is greater than or equal to the predetermined scale index threshold, a person is included in the extracted time series region. and a person determination unit (15) that determines that the person is present.

According to the person detection device (10) including the person determination unit (15), the value of the irregularity index calculated by the person determination unit (15) is greater than or equal to the predetermined irregularity index threshold, and the person determination unit (15) performs the calculation. If the value of the scale index is greater than or equal to the predetermined scale index threshold, it is determined that a person is included in the extracted time-series region. It is possible to distinguish between objects such as electrical equipment. Therefore, it is possible to prevent an object such as an electronic device or an electric device that does not move from being mistakenly detected as a person.

(Additional note 2)
The person detection device according to Supplementary note 2 is the person detection device described in Supplementary note 1, wherein the partial area includes a part of a high temperature area that is higher than a predetermined high temperature threshold and a low temperature area that is lower than the high temperature threshold. include.

"Part" of the high-temperature region means a part of the high-temperature region, not all of it. According to the person detection device of Appendix 2, since the partial region includes a part of the high temperature region that is higher than the predetermined high temperature threshold and a low temperature region that is lower than the high temperature threshold, it is possible to better detect fluctuations in the value of the fluctuation index. can be observed. This makes it easier to capture the movement of a person.

(Additional note 3)
The person detection device according to Supplementary Note 3 is the person detection device described in Supplementary Note 2, in which the person determination unit is configured to detect a person in which the value of the calculated irregularity index is equal to or greater than a predetermined irregularity index threshold. If the value of the calculated scale index is equal to or greater than a predetermined scale index threshold, it is determined that a part of the high temperature region in the extracted time series region is a person.

(Additional note 4)
The person detection device of appendix 4 is the person detection device described in any one of appendices 1 to 3, in which the variation index is the temperature center of gravity of the partial region.

(Appendix 5)
The person detection device according to appendix 5 is the person detection device described in any one of appendices 1 to 4, and the value of the irregularity index is the p value of the Cox-Start test.

(Appendix 6)
The person detection device according to Appendix 6 is the person detection device described in any one of Appendixes 1 to 5, and the value of the scale index is a value obtained by linear regression of the calculated value of the fluctuation index, and the calculated value. It is the value of the standard deviation of the difference from the value of the fluctuation index.

(Appendix 7)
The person detection device according to appendix 7 is the person detection device described in any one of appendices 1 to 6, and the person detection device according to appendix 7 refers to the learned model that has learned whether or not the imaged object is a person. The person determination unit further includes a learning result utilization unit (14) that acquires the degree of personhood of objects included in a plurality of thermal images or heat maps, and the person determination unit determines whether the value of the calculated irregularity index is a predetermined irregularity. Extraction is performed when the regularity index is greater than or equal to the regularity index threshold, the calculated scale index value is greater than or equal to the predetermined scale index threshold, and the obtained personhood is greater than or equal to the predetermined personhood threshold. It is determined that the object imaged in the time-series area is a person.

(Appendix 8)
The person detection method in Appendix 8 is person detection performed by a person detection device including a time-series region extraction unit (11), a fluctuation index value calculation unit (12), a fluctuation characteristic calculation unit (13), and a person determination unit (15). The method includes a step in which the time series region extraction unit (11) extracts time series of partial regions of the plurality of thermal images or heat maps as time series regions from a plurality of thermal images or heat maps at different times. (ST1), and a step (ST1) in which the variation index value calculation unit (12) calculates the value of a variation index that quantifies the temperature distribution or the label distribution for each partial region included in the extracted time series region. ST2), and the fluctuation characteristic calculation unit (13) calculates, for the calculated fluctuation index value, the value of the irregularity index indicating the irregularity of fluctuation and the value of the scale index indicating the magnitude of the fluctuation. In the step of calculating (ST3), the person determining unit (15) determines that the value of the calculated irregularity index is greater than or equal to a predetermined irregularity index threshold, and the value of the calculated scale index is a predetermined value. and a step (ST5) of determining that a person is included in the extracted time-series region if the size index is equal to or greater than a predetermined scale index threshold.

(Appendix 9)
The person detection program according to appendix 9 includes a time-series region extraction function that extracts, as a time-series region, a time-series of partial regions of the plurality of thermal images or heat maps from a plurality of thermal images or heat maps at different times; A variation index value calculation function that calculates the value of a variation index that quantifies the temperature distribution or label distribution for each partial region included in the time series region, and a variation index value calculation function that calculates the value of a variation index that quantifies the distribution of temperature or label. A fluctuation characteristic calculation function that calculates the value of the irregularity index that indicates the nature of the fluctuation and the value of the scale index that represents the magnitude of the fluctuation, and the irregularity index threshold that the calculated irregularity index value is set in advance. and a person determination function that determines that a person is included in the extracted time series region when the value of the calculated scale index is equal to or greater than a predetermined scale index threshold. Let it run.

Note that it is possible to combine the embodiments, or to modify or omit each embodiment as appropriate.

The person detection technology of the present disclosure can detect a person from a low-resolution thermal image more accurately than before, so it can be incorporated into a device equipped with a low-resolution camera, such as an air conditioner.

Reference Signs List 10 person detection device, 11 time series region extraction unit, 12 variation index value calculation unit, 13 variation characteristic calculation unit, 14 learning result utilization unit, 15 person determination unit, 20 imaging device, 100a processing circuit, 100b processor, 100c memory.

Claims (9)

a time-series region extraction unit that extracts a time-series partial region of the plurality of thermal images or heatmaps as a time-series region from a plurality of thermal images or heatmaps at different times;
a fluctuation index value calculation unit that calculates the value of a fluctuation index that quantifies the temperature distribution or the label distribution for each partial region included in the extracted time series region;
a fluctuation characteristic calculation unit that calculates a value of an irregularity index indicating the irregularity of the fluctuation and a value of a scale index indicating the magnitude of the fluctuation for the calculated value of the fluctuation index;
A time series extracted when the value of the calculated irregularity index is equal to or greater than the predetermined irregularity index threshold, and the value of the calculated scale index is equal to or greater than the predetermined scale index threshold. a person determination unit that determines that a person is included in the area;
A person detection device equipped with. The person detection device according to claim 1, wherein the partial area includes a part of a high temperature area that is higher than a predetermined high temperature threshold and a low temperature area that is lower than the high temperature threshold. The person determination unit determines whether the calculated irregularity index value is equal to or greater than a predetermined irregularity index threshold value, and the calculated scale index value is equal to or greater than a predetermined scale index threshold value. 3. The person detection device according to claim 2, wherein a part of the high temperature area in the extracted time series area is determined to be a person. The person detection device according to claim 1, wherein the variation index is a temperature center of gravity of a partial region. The person detection device according to claim 1, wherein the value of the irregularity index is a p value of Cox-Start test. 2. The person detection device according to claim 1, wherein the value of the scale index is a value of a standard deviation of a difference between a value obtained by linear regression of the calculated value of the fluctuation index and the calculated value of the fluctuation index. further comprising a learning result utilization unit that obtains the degree of personhood of the object included in the plurality of thermal images or heat maps by referring to a trained model that has learned whether the imaged object is a person;
The person determination unit has a calculated irregularity index value that is equal to or greater than a predetermined irregularity index threshold, and a calculated scale index value that is equal to or greater than a predetermined scale index threshold, and determining that the object imaged in the extracted time-series region is a person when the obtained person degree is equal to or greater than a predetermined person degree threshold;
A person detection device according to any one of claims 1 to 6. A person detection method performed by a person detection device including a time-series region extraction unit, a variation index value calculation unit, a variation characteristic calculation unit, and a person determination unit,
the time-series region extracting unit extracting time-series partial regions of the plurality of thermal images or heat maps as time-series regions from a plurality of thermal images or heat maps at different times;
a step in which the variation index value calculation unit calculates a variation index value that quantifies the temperature distribution or the label distribution for each partial region included in the extracted time series region;
The fluctuation characteristic calculation unit calculates, for the calculated fluctuation index value, a value of an irregularity index indicating irregularity of fluctuation and a value of a scale index indicating the magnitude of the fluctuation;
When the person determination unit determines that the calculated irregularity index value is equal to or greater than a predetermined irregularity index threshold, and the calculated scale index value is equal to or greater than a predetermined scale index threshold, , determining that a person is included in the extracted time series region;
A person detection method comprising: a time series region extraction function that extracts a time series of partial regions of the plurality of thermal images or heat maps as a time series region from a plurality of thermal images or heat maps at different times;
a fluctuation index value calculation function that calculates the value of a fluctuation index that quantifies the temperature distribution or label distribution for each partial region included in the extracted time series region;
a fluctuation characteristic calculation function that calculates, for the calculated fluctuation index value, an irregularity index value indicating the irregularity of the fluctuation and a scale index value indicating the magnitude of the fluctuation;
A time series extracted when the value of the calculated irregularity index is equal to or greater than the predetermined irregularity index threshold, and the value of the calculated scale index is equal to or greater than the predetermined scale index threshold. A person determination function that determines that a person is included in the area;
A person detection program that runs a computer.

Images