Detector

Abstract

To improve the accuracy of detecting dirt on an object by using an image taken of the object when there is a possibility that an obstruction will be captured in the image of the object.SOLUTION: If an obstruction is detected in at least one of two images taken of an object at different times, a detector 20 detects the amount of dirt on the object by excluding the area occupied by the obstruction from both images and then comparing the difference in the remaining areas.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a technique for detecting dirt on an object. [Background technology]

[0002] There is a known technology for detecting dirt on an object by image processing. For example, Patent Document 1 describes a technology for detecting dirt on a camera lens by calculating the difference between a current image and the previous image captured by the camera. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2010-273014 A Summary of the Invention [Problem to be solved by the invention]

[0004] However, with the technology described in Patent Document 1, if an obstructing object such as a person is captured in an image captured by the camera, the obstructing object may be erroneously detected as dirt.

[0005] The present invention aims to improve the accuracy of detecting dirt on an object using an image captured of the object when there is a possibility that an obstructing object is captured in the image. [Means for solving the problem]

[0006] One aspect of the present invention provides a detection device that, when an obstruction is detected in at least one of two images taken of an object at different times, removes the portion of the obstruction from both images and detects the amount of dirt on the object from the difference in the remaining area.

[0007] The detection device may detect the dirty portion of the object from the difference.

[0008] The sensing device may provide an indication if the amount exceeds a threshold.

[0009] The detection device may detect the obstruction using a trained model that is machine-learned on a predetermined obstruction.

[0010] The difference may be a difference in luminance of the regions.

[0011] The detection device may measure a distance to an object being photographed, including the target object, and detect the obstruction based on the measured distance.

[0012] The detection device may measure the distance to the object based on a plurality of images each of which is taken from different positions with a time difference less than a threshold value. Effect of the Invention

[0013] According to the present invention, when there is a possibility that an obstructing object is captured in an image of an object, the accuracy of detecting dirt on the object using the image is improved. [Brief description of the drawings]

[0014] [Figure 1] FIG. 1 is a diagram illustrating an example of the configuration of a dirt detection system. [Diagram 2] FIG. 2 is a diagram illustrating an example of capturing an image of an object by a camera. [Diagram 3] 4 is a flowchart illustrating a dirt detection process. [Figure 4] 10 is a flowchart illustrating details of an accidental soiling detection process. [Diagram 5] 6A to 6C are diagrams illustrating an example of image processing for detecting sudden soiling. [Figure 6] 6A to 6C are diagrams illustrating an example of image processing for detecting sudden soiling. [Figure 7] 6A to 6C are diagrams illustrating an example of image processing for detecting sudden soiling. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] <Configuration> FIG. 1 is a diagram illustrating an example of the configuration of a dirt detection system 1. The dirt detection system 1 is installed, for example, in a station, detects dirt on objects to be cleaned such as floors, walls, and windows, and prompts cleaning work. Note that the location where the dirt detection system 1 is installed is not limited to a station, and it may be a place other than a station, such as a commercial facility. The dirt detection system 1 includes multiple cameras 10 and a detection device 20. The multiple cameras 10 and the detection device 20 are connected to each other so as to be able to communicate with each other via a wired or wireless network 2.

[0016] FIG. 2 is a diagram illustrating an example of the camera 10 photographing an object to be cleaned. The camera 10 is a fixed camera installed in a position where it can photograph the object to be cleaned, and photographs an image of the object to be cleaned. The image obtained by photographing the object by the camera 10 is digital image data. In the example shown in FIG. 2(a), the camera 10 is installed near the ceiling of the station, and photographs an image of a photographing range 103 including a part of a floor 101 and a part of a wall 102. FIG. 2(b) is a diagram illustrating an image 100 photographed by the camera 10. As shown in FIG. 2(b), a part of the floor 101 and a part of a wall 102 are photographed in the image 100. The camera 10 photographs images at a predetermined time interval, such as 5 minutes, and sequentially transmits the photographed images to the detection device 20 with the photographed date and time added to the photographed images. The camera 10 does not necessarily have to photograph images at the same time interval, and may photograph images at different time intervals. Moreover, information such as a serial number indicating the order in which the images were photographed may be added to the images instead of the photographed date and time.

[0017] Returning to FIG. 1, the detection device 20 performs a dirt detection process to detect dirt on the object to be cleaned by applying image processing to the image captured by the camera 10. This dirt detection process includes an accumulated dirt detection process to detect gradually accumulating dirt, and an accidental dirt detection process to detect accidental dirt such as vomit or littering. The detection device 20 is a computer such as a server. The detection device 20 includes a processor 21, a communication interface 22, a memory 23, and a display unit 24. These components are connected via a bus.

[0018] The processor 21 controls and performs calculations for each part of the detection device 20. The processor 21 also has a timekeeping function. An example of the processor 21 is a central processing unit (CPU). The communication interface 22 communicates with the camera 10 via the network 2 under the control of the processor 21. The communication interface 22 receives images from the camera 10. The memory 23 stores a dirt detection program 231 and an image database 232. The dirt detection program 231 is a program for causing the detection device 20 to execute a dirt detection process. The image database 232 stores images received from the camera 10. Examples of the memory 23 include a read only memory (ROM), a random access memory (RAM), a hard disk drive (HDD), a solid state drive (SSD), and a flash memory. The display unit 24 displays a warning message encouraging a cleaning operation in response to detection of dirt on the cleaning target under the control of the processor 21. Examples of the display unit 24 include a liquid crystal display and an organic EL (electroluminescence) display.

[0019] The processor 21 executes a dirt detection program 231 stored in the memory 23 to perform calculations or control each part of the detection device 20, thereby functioning as an acquisition part 211, a switching part 212, an accumulated dirt detection part 213, an object detection part 214, an accidental dirt detection part 215, a determination part 216, and a notification part 217. These functions are realized by software working in cooperation with hardware resources.

[0020] The acquisition unit 211 receives images from the camera 10 via the communication interface 22, associates the received images with the shooting dates and times, and stores them in the image database 232. The acquisition unit 211 also reads and acquires a target image and a reference image that were shot at different times by the camera 10 from the image database 232. The target image is, for example, the latest image shot by the camera 10. The reference image is, for example, the image shot by the camera 10 immediately before the latest image.

[0021] The switching unit 212 switches between the accumulated dirt detection process and the sudden dirt detection process depending on the time when the dirt detection process is executed. The accumulated dirt detection process is performed using images taken during a specified time when people are not captured in the images of the object to be cleaned, such as at night when the station is closed. On the other hand, the sudden dirt detection process is performed in real time using images taken outside the specified time, that is, when the station is open, and other times when people may be captured in the images of the object to be cleaned.

[0022] The accumulated dirt detection unit 213 performs an accumulated dirt detection process by performing image processing on the target image captured within a specified time. More specifically, the accumulated dirt detection unit 213 detects accumulated dirt based on the difference between the brightness value of the target image and a reference value. The reference value is, for example, the brightness value of an image of the object to be cleaned in an initial state.

[0023] The object detection unit 214 detects a person from a target image taken outside the specified time and a reference image. There is a possibility that a person is captured in an image taken outside the specified time. The object detection unit 214 detects a person captured in the image. More specifically, the object detection unit 214 detects a person using AI (Artificial Intelligence). For example, the object detection unit 214 detects a person using a machine-learned model that has been machine-learned on a person. The object detection unit 214 may also detect a person by deep learning. A person is an example of an obstructing object according to the present invention.

[0024] In this embodiment, the object detection unit 214 detects a person, but an obstructing object other than a person may be detected. The obstructing object detected by the object detection unit 214 may be an object other than dirt that may appear as a difference between a target image and a reference image. More specifically, the obstructing object is an object that may suddenly appear in the shooting range of the camera 10. Examples of the obstructing object include moving objects such as a dolly or a train, and objects that are temporarily placed. When detecting an obstructing object using a machine learning model, the machine learning model learns predetermined obstructing objects by machine learning.

[0025] The sudden stain detection unit 215 performs the sudden stain detection process by performing image processing on the target image and the reference image taken outside the specified time. More specifically, the sudden stain detection unit 215 excludes the human area, which is the part occupied by a person, from the target image and the reference image taken outside the specified time, and detects the sudden stain on the cleaning object from the difference between the luminance value of the target image and the luminance value of the reference image. When a person is captured in at least one of the target image and the reference image, not only the stain on the cleaning object but also the person appears as the difference between the target image and the reference image. Therefore, by excluding the human area from the target image and the reference image, it is possible to detect only the stain on the cleaning object from the difference between the target image and the reference image. In this embodiment, the sudden stain detection unit 215 detects the presence or absence of stain, but detecting stain means detecting an amount of stain equal to or greater than the minimum amount that can be detected. In addition, in this embodiment, the luminance value is used in the image processing for detecting stain, but instead of the luminance value, a saturation value, a gradation value, a brightness value, or a pixel value may be used.

[0026] The determination unit 216 determines whether or not cleaning work is necessary based on the detection result of the accumulated dirt detection unit 213 or the sudden dirt detection unit 215. For example, the determination unit 216 determines that cleaning work is necessary when dirt is detected by the accumulated dirt detection unit 213 or the sudden dirt detection unit 215, and determines that cleaning work is not necessary when dirt is not detected.

[0027] When the determination unit 216 determines that cleaning work is necessary, the notification unit 217 issues a notification to prompt the user to perform cleaning work. More specifically, when the determination unit 216 determines that cleaning work is necessary, the notification unit 217 causes the display unit 24 to display a warning message prompting the user to perform cleaning work. Note that the notification to prompt the user to perform cleaning work is not limited to displaying the warning message, and may be a sound output from a speaker prompting the user to perform cleaning work, or a message prompting the user to perform cleaning work may be sent to a mobile terminal carried by the cleaner.

[0028] <Operation> 3 is a flowchart illustrating a dirt detection process performed by the detection device 20. Each camera 10 captures an image of the object to be cleaned at a predetermined time interval, adds the shooting date and time to the captured image as metadata, and transmits the image to the detection device 20. The dirt detection process is started at a predetermined time interval when an image is received from the camera 10.

[0029] In step S11, the acquisition unit 211 receives images transmitted from each camera 10 via the communication interface 22, and stores the received images in the image database 232 in association with the image capture date and time.

[0030] In step S12, the switching unit 212 determines whether the current time is within the designated time. If the current time is within the designated time (YES in step S12), the switching unit 212 instructs execution of the accumulated dirt detection process. As a result, the process proceeds to step S13.

[0031] In step S13, the accumulated dirt detection unit 213 executes the accumulated dirt detection process using the image received in step S11. More specifically, the accumulated dirt detection unit 213 uses the received image as a target image and calculates the difference between the brightness value of the target image and a reference value. As described above, the designated time is a time when people are not captured in the image of the object to be cleaned, such as at night when the station is closed, so that no people are captured in the target image captured during the designated time. Therefore, if the difference between the brightness value of the target image and the reference value exceeds the threshold value, it indicates that accumulated dirt is attached to the object to be cleaned. Therefore, the accumulated dirt detection unit 213 detects accumulated dirt when this difference exceeds the threshold value, and does not detect accumulated dirt when the difference is equal to or less than the threshold value.

[0032] On the other hand, in the above-mentioned step S12, if it is determined that the current time is outside the designated time (determination in step S12 is NO), the switching unit 212 instructs execution of the accidental stain detection process. As a result, the process proceeds to step S14. In step S14, the accidental stain detection process is executed. Details of the accidental stain detection process will be described later.

[0033] In step S15, the determination unit 216 determines whether or not cleaning work is necessary based on the results of the accumulated dirt detection process or the sudden dirt detection process described above. For example, the determination unit 216 determines that cleaning work is necessary if dirt is detected in the accumulated dirt detection process or the sudden dirt detection process described above, and determines that cleaning work is unnecessary if dirt is not detected. If it is determined that cleaning work is necessary (YES in step S15), the process proceeds to step S16.

[0034] In step S16, notification unit 217 causes display unit 24 to display a warning message prompting cleaning work, and the dirt detection process ends. On the other hand, if it is determined that cleaning work is not necessary (NO in step S15), the dirt detection process ends without performing the process in step S16.

[0035] Fig. 4 is a flowchart illustrating the details of the accidental stain detection process in step S14 described above. Figs. 5 to 7 are diagrams illustrating an example of image processing for detecting accidental stain.

[0036] In step S21, the acquisition unit 211 reads and acquires the target image and the reference image from the image database 232 with reference to the photographing date and time. Here, an example will be described in which the predetermined time interval is 5 minutes. If the current time is 10:05, in the example shown in FIG. 5, the acquisition unit 211 first reads the image 100B photographed by the camera 10 at 10:05 as the target image from the image database 232. Next, the acquisition unit 211 reads the image 100A photographed by the same camera 10 at 10:00, which is 5 minutes before the current time, from the image database 232 as the reference image. In this example, a person is photographed in each of the images 100A and 100B. Also, since dirt occurred on the floor 101 photographed in the images 100A and 100B after 10:00 and before 10:05, dirt that is not present in the image 100A is photographed in the image 100B.

[0037] In step S22, the object detection unit 214 performs human detection processing on each of the target image and the reference image acquired in step S21. In the example shown in Fig. 5, the image 100A includes a human region 105, and the image 100B includes a human region 106. Therefore, the object detection unit 214 detects the human region 105 from the image 100A, and detects the human region 106 from the image 100B.

[0038] In step S23, the accidental stain detection unit 215 judges whether or not a person is captured in at least one of the target image and the reference image based on the result of the person detection process in step S22. For example, if a human region is detected in at least one of the target image and the reference image, the accidental stain detection unit 215 judges that a person is captured in at least one of the target image and the reference image, and if a human region is not detected in either the target image or the reference image, the accidental stain detection unit 215 judges that a person is not captured in either the target image or the reference image. If a person is not captured in either the target image or the reference image (NO in step S23), the process proceeds to step S24.

[0039] In step S24, the accidental stain detection unit 215 creates a difference image between the target image and the reference image acquired in step S21. More specifically, the accidental stain detection unit 215 creates the difference image by cutting out a difference region where the value obtained by subtracting the brightness value of the reference image from the brightness value of the target image is other than 0. Since no person is captured in the target image and the reference image used in step S24, the presence of a difference between the target image and the reference image indicates that there is dirt on the object to be cleaned. Therefore, the accidental stain detection unit 215 detects accidental stains when the difference region between the target image and the reference image is included in the difference image. On the other hand, the accidental stain detection unit 215 does not detect accidental stains when the difference region between the target image and the reference image is not included in the difference image and only the background region is included.

[0040] On the other hand, if it is determined in step S23 that a person is included in at least one of the target image and the reference image (YES in step S23), the process proceeds to step S25. In step S25, the accidental stain detection unit 215 creates a mask image by masking the human region detected in step S22 for each of the target image and the reference image acquired in step S21. That is, the accidental stain detection unit 215 excludes the human region from both the target image and the reference image. In the example shown in FIG. 5, the accidental stain detection unit 215 creates a mask image 110B by converting the luminance value of the human region 106 included in the image 100B to 0 and masking it, and removes the region other than the masked human region 106. In addition, the accidental stain detection unit 215 creates a mask image 110A by converting the luminance value of the human region 105 in the image 100A to 0 and masking it, and removes the region other than the masked human region 105.

[0041] In step S26, the accidental stain detection unit 215 creates a composite mask image by adding together the mask image of the target image created in step S25 and the mask image of the reference image. In the example shown in Fig. 5, the accidental stain detection unit 215 creates a composite mask image 120 by adding up the luminance value of the mask image 110A and the luminance value of the mask image 110B for each pixel. The composite mask image 120 includes both the human region 106 included in the mask image 110B and the human region 105 included in the image 100A.

[0042] In step S27, the accidental stain detection unit 215 colors the human regions included in the composite mask image created in step S26 with a first color. In the example shown in Fig. 7, the accidental stain detection unit 215 converts the luminance value of the human regions 105 and 106 in the composite mask image 120, i.e., a luminance value of 0, into a luminance value indicating the first color. As a result, the color of the human regions 105 and 106 becomes the first color.

[0043] In step S28, the accidental stain detection unit 215 creates a difference image between the target image and the reference image acquired in step S21. In the example shown in Fig. 6, the accidental stain detection unit 215 creates a difference image 130 by cutting out a difference region where the value obtained by subtracting the luminance value of image 100A from the luminance value of image 100B is other than 0. The difference image 130 includes a human region 105, a human region 106, and a stain region 107 as difference regions. The luminance value of the background region of the difference image 130 is 0, and the luminance values ​​of these difference regions are values ​​other than 0.

[0044] In step S29, the accidental stain detection unit 215 colors the difference region of the difference image with a second color. In order to easily distinguish between the human region and the stain region, it is preferable that the second color is different from the first color described above. In the example shown in FIG. 7, the accidental stain detection unit 215 converts the luminance values ​​of the human region 105, the human region 106, and the stain region 107 included in the difference image 130, i.e., luminance values ​​other than 0, into luminance values ​​indicating the second color. As a result, the colors of the human region 105, the human region 106, and the stain region 107 become the second color.

[0045] In step S30, the accidental stain detection unit 215 creates a composite image by multiplying the composite mask image colored in the first color onto the difference image colored in the second color. In the example shown in FIG. 7, the accidental stain detection unit 215 creates a composite image 140 by multiplying the luminance value of the difference image 130 by the luminance value of the composite mask image 120 for each pixel. In the composite image 140, the human region 105 and the human region 106 have different colors from the stain region 107, so the accidental stain detection unit 215 can distinguish and extract the human regions 105 and 106 from the stain 107 based on the luminance values ​​of these regions. The stain region 107 is the difference between the remaining region after excluding the human regions 105 and 106 from the images 100A and 100B. The presence of a difference in the remaining region after excluding the human regions 105 and 106 indicates that the object to be cleaned has stains. Therefore, the accidental stain detection unit 215 detects accidental stain when the stained area 107 is extracted from the composite image 140. On the other hand, the accidental stain detection unit 215 does not detect accidental stain when the stained area is not extracted from the composite image 140. When the processing of step S30 is completed, the accidental stain detection processing ends.

[0046] In the above-mentioned step S16, a warning message is displayed to prompt the user to perform cleaning, and if the user performs cleaning of the dirt, the image captured after the cleaning will not include the dirty area. Even if dirt is removed by cleaning, a difference will occur between the target image and the reference image, but in this case, it is preferable not to detect the dirt or to notify the user to perform the cleaning. Therefore, when dirt is removed from the object to be cleaned, the accidental dirt detection unit 215 converts the brightness value of the dirty area so that the dirty area is not extracted from the difference image.

[0047] Here, as shown in FIG. 6, it is assumed that after the image 100B is taken at 10:05, dirt on the object to be cleaned is removed by cleaning work, and then an image of the object to be cleaned is taken at 10:10. In this case, in the accidental dirt detection process, if a difference image is created using the image taken at 10:10 as the target image and the image 100B taken at 10:05 as the reference image, the brightness value of the area corresponding to the dirty area 107 included in the image 100B in this difference image becomes a negative value. The fact that the brightness value of the area corresponding to the dirty area 107 in the difference image becomes a negative value indicates that the dirt has been removed by cleaning work. Therefore, the accidental dirt detection unit 215 converts the brightness value of the area corresponding to the dirty area 107 in the difference image, i.e., the negative brightness value, to 0, which is the same as the brightness value of the background area. Since the dirty area is no longer extracted due to this conversion of the brightness value, the detection of dirt and the notification to prompt the user to perform cleaning are not performed.

[0048] According to the above-mentioned embodiment, the area occupied by an obstructing object such as a person is excluded from the target image and the reference image, and dirt on the cleaning object is detected from the difference between the target image and the reference image. Therefore, even if an obstructing object such as a person is captured in the image of the cleaning object, the obstructing object is not erroneously detected as dirt. Therefore, when detecting dirt on the cleaning object using such an image, the accuracy of detecting dirt is improved. Furthermore, according to the above-mentioned embodiment, it is possible to detect dirt in real time during a time period when a person may be captured in an image of the cleaning object, such as when the station is open, so that it is possible to detect sudden dirt that occurs during such a time period and quickly clean the dirt. If it is possible to detect dirt in real time and to notify the user to perform cleaning work, the cleaning work only needs to be performed when dirt is generated on the target object, and the efficiency of the cleaning work is improved compared to the case where cleaning work is performed periodically. Furthermore, since the accumulated dirt detection process is performed within the specified time and the sudden dirt detection process is performed outside the specified time, these dirt detection processes make it possible to detect a wide variety of dirt. Furthermore, the accuracy of detecting people is improved by detecting people using AI such as machine learning.

[0049] <Modification> The above is a description of the embodiment, but the contents of this embodiment may be modified as follows. In addition, the following modifications may be combined.

[0050] In the above-described embodiment, the accidental stain detection unit 215 may further calculate the difference between the brightness value of the target image and the brightness value of the reference image, and detect the amount of stain according to the calculated difference in brightness value. For example, the accidental stain detection unit 215 detects that the amount of stain is greater as the calculated difference in brightness value is greater. Then, when the accidental stain detection unit 215 detects an amount of stain that exceeds a threshold, that is, when a portion with a lot of stains is detected, the determination unit 216 may determine that cleaning work is necessary, and the notification unit 217 may display a warning message prompting cleaning work on the display unit 24. In other words, when the amount of stain detected by the accidental stain detection unit 215 is equal to or less than the threshold, the determination unit 216 may determine that cleaning work is unnecessary, and the notification unit 217 may not notify the user to perform cleaning work. According to this modified example, a notification prompting cleaning work can be performed only when the amount of stain on the cleaning target is large. In other words, even if the cleaning target is dirty, when the amount of stain is small, a notification prompting cleaning work can be notified.

[0051] In the above-described embodiment, if the dirt in the image happens to look like a person, the object detection unit 214 may erroneously detect the dirt in the image as a person. Therefore, in order to improve the accuracy of detecting an obstructing object such as a person, the object detection unit 214 may detect the obstructing object by taking into account the distance from the camera 10 to the photographed object. Methods for measuring this distance include, for example, a method using triangulation with a stereo camera and a method using a distance measuring device of the ToF (Time Of Flight) method, but other known methods may also be used. In the triangulation method, the camera 10 is a stereo camera, and the stereo camera takes multiple images from different positions at the same time and transmits them to the detection device 20. Note that the timing at which the stereo camera takes multiple images does not necessarily have to be simultaneous, and there may be a time difference less than a threshold. The detection device 20 measures the distance to the photographed object using the images received from the camera 10 using the principle of triangulation, and detects the obstructing object based on the measured distance. More specifically, the distance from the camera 10 to the object to be cleaned is determined in advance and stored in the memory 23 of the detection device 20. If the distance to the object position of the human candidate indicated by the human area detected using AI is a predetermined distance, the human candidate may not be a person but dirt on the cleaning target. Therefore, the object detection unit 214 does not detect the human candidate as a person if the distance from the camera 10 to the object position of the human candidate is a predetermined distance, and detects the human candidate as a person if the distance is other than the predetermined distance. This modification improves the accuracy of detecting people.

[0052] In the above-described embodiment, the configuration of the dirt detection system 1 is not limited to the above-described example. The dirt detection system 1 may be configured to include one or more of the above-described devices, or may be configured without including some of the devices. The detection device 20 may also be configured to include one or more of the above-described components, or may be configured without including some of the components.

[0053] In the above-described embodiment, the processor 21 is not limited to a CPU, and may include a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), or other programmable logic devices. At least a part of the functions of the detection device 20 may be realized by a circuit such as a digital signal processor (DSP), an application specific integrated circuit (ASIC), a programmable logic device (PLD), or a field programmable gate array (FPGA).

[0054] In the above-described embodiment, the operations of the dirt detection system 1 and the detection device 20 are not limited to the above-described examples. The order of the processing steps of the dirt detection system 1 and the detection device 20 may be changed or some processing steps may be omitted, as long as there is no contradiction.

[0055] Another aspect of the present invention may provide a method having processing steps performed in the dirt detection system 1. Furthermore, yet another aspect of the present invention may provide a program executed in the detection device 20. This program may be provided by being stored in a computer-readable recording medium, or may be provided by downloading via the Internet or the like. [Explanation of symbols]

[0056] 1: detection system, 2: network, 10: camera, 20: detection device, 21: processor, 22: communication interface, 23: memory, 24: display unit, 211: acquisition unit, 212: switching unit, 213: accumulated dirt detection unit, 214: object detection unit, 215: sudden dirt detection unit, 216: determination unit, 217: notification unit

Claims

1. This detection device detects the amount of dirt on an object from the difference in the remaining area by excluding the portion of the object that is occupied by the obstruction from both images taken of the object at different times when an obstruction is detected in at least one of the two images.

2. The part of the object that is more contaminated is detected based on the difference. The detection device of claim 1 .

3. Notify if the amount exceeds a threshold The detection device of claim 1 .

4. Detecting a predetermined obstruction using a trained model that has been machine-learned for the predetermined obstruction The detection device of claim 1 .

5. The difference is the difference in luminance of the region. The detection device of claim 1 .

6. Measure the distance to the photographed object, including the target object, and detect the obstruction based on the measured distance. The detection device of claim 1 .

7. Measure the distance to the object based on a plurality of images taken from different positions with a time difference less than a threshold value.

7. The detection device of claim 6.

Images