TWI771672B - Image monitoring apparatus and method - Google Patents

Abstract

An image monitoring apparatus including an image sensing module and a processor is provided. The image sensing module is configured to obtain an invisible light dynamic image of an objective scene. The invisible light dynamic image includes a plurality of frames. The processor is configured to perform operations according to at least one frame of the invisible light dynamic image to determine a status of at least one live body corresponding to the objective scene to be one of a plurality of status types and at least one status valid region of the invisible light dynamic image, and set a scene of each pixel of the at least one status valid region to be one of a plurality of scene types corresponding to the status type of the at least one live body. An image monitoring method is also provided.

Description

Image monitoring device and method

The present invention relates to a monitoring device and method, and more particularly, to an image monitoring device and method.

With the advancement of medical technology, the average life expectancy of human beings has been extended, so there is a need for silver hair health care. In addition, in the phenomenon of silver-haired people living at home, the number of silver-haired people living alone accounts for a certain proportion, and institutional and community care manpower is limited. Therefore, the world is developing home care services with the help of technology.

The main accidental movement injuries of silver-haired people are the behavior of getting out of bed in the bedroom, abnormal abnormal movements, and the ground is slippery. Therefore, prevention and immediate treatment are important needs of home security. For example, silver-haired people get up at night and fall, only to be found the next morning. Another example is a silver-haired person who is unwell in bed and cannot seek help. Therefore, real-time notification of abnormal movements is an urgent need.

Most of the existing care systems are based on wearable sensing devices or pressure pads. However, the sensors need to be worn for a long time, and the elderly are not willing to wear them or they will be removed by themselves. In addition, the limited range of pressure pads makes it impossible to sense abnormal fall events at any time. On the other hand, the current artificial intelligence (AI) recognition technology has a high accuracy rate in motion recognition, but it uses general images for recognition. General images refer to the user's facial features, clothing, or body surface and other privacy content. Therefore, the care recipients will feel that their privacy has been violated and will reduce their willingness to install.

The present invention provides an image monitoring device and method, which can provide good and effective safety monitoring under the condition of using the low-sensitivity image of the care recipient.

An embodiment of the present invention provides an image monitoring device including an image sensing module and a processor. The image sensing module is configured to obtain a non-visible light dynamic image of a target scene, and the non-visible light dynamic image includes several image frames. The processor is configured to perform an operation according to at least one image frame of the non-visible light dynamic image to determine that a state corresponding to at least one living body in the target scene is one of several state categories and at least one state valid block of the non-visible light dynamic image , and setting a scene of each pixel of the at least one state valid block as one of a plurality of scene categories according to the at least one active state class.

An embodiment of the present invention provides an image monitoring method, including: obtaining a non-visible light dynamic image of a target scene; performing an operation according to at least one image frame of the non-visible light dynamic image to determine a state corresponding to at least one living body in the target scene One of several state classes and at least one state valid block of the non-visible light dynamic image, and setting a scene of each pixel of the at least one state valid block according to the state class of the at least one living body to be one of the several scene classes .

In the image monitoring device and method of the embodiments of the present invention, non-visible light dynamic images are used to identify the living body, the state type and the state valid block, and the scene of the pixels in the state valid block is set according to the state type of the living body. One of the scene categories. Therefore, the image monitoring apparatus and method in the embodiments of the present invention can utilize the low-sensitivity image of the care recipient to perform good and effective security monitoring, so as to maintain the privacy of the care recipient.

FIG. 1 is a schematic diagram of an image monitoring apparatus according to an embodiment of the present invention, and FIG. 2 shows a non-visible light dynamic image obtained by the image monitoring apparatus of FIG. 1 . Referring to FIG. 1 and FIG. 2 , the image monitoring device 100 of this embodiment includes an image sensing module 110 and a processor 120 . The image sensing module 110 is configured to obtain a non-visible light dynamic image of a target scene, and the non-visible light dynamic image includes several image frames (one of which is shown in FIG. 2 ), that is, the non-visible light dynamic image. The visible light dynamic image is composed of the several image frames that are imaged at different time points respectively. In this embodiment, the non-visible light dynamic image is a thermal image, and the image sensing module 110 can be a thermal image camera for detecting the thermal image. However, in other embodiments, the non-visible light dynamic image can also be a radio frequency echo image or an ultrasonic image, and the image sensing module 110 can be an ultrasonic transceiver or a radio frequency electromagnetic wave transceiver.

The processor 120 is configured to perform the following steps. First, the processor 120 performs an operation according to at least one image frame of the non-visible light dynamic image (eg, the image frame shown in FIG. 2 ), and determines that the state corresponding to at least one living body 60 in the target scene is one of several state categories and non- At least one active block A1 of the visible light dynamic image. Next, according to the state class of the at least one living body 60 , a scene of each pixel of the at least one state valid block A1 is set as one of several scene classes. For example, the living body 60 is a human body, and the state class includes at least one of standing, sitting, lying down, crawling, and undefined, and the state class of the living body 60 shown in FIG. 2 is, for example, lying. Furthermore, these scene categories include, for example, at least one of floor 52, bed 54, seat 56, and undefined categories.

In the embodiment shown in FIGS. 1 and 2 , when the processor 120 determines that the state type of the at least one living body 60 is station, the processor 120 sets the scene of each pixel of the at least one state valid block A1 as the floor. When the processor 120 determines that the state type of the at least one living body 60 is sitting, the processor 120 sets the scene of each pixel of the at least one state valid block A1 as a seat. When the processor 120 determines that the state type of the at least one living body 60 is lying down, the processor 120 sets the scene of each pixel of the at least one state valid block A1 as a bed. Taking the living body 60 shown in FIG. 2 as an example, the processor 120 determines that its state type is lying down and the corresponding scene type is a bed, so the processor 120 further sets the scene of each pixel in the state valid block A1 as a bed .

3A, 3B, and 3C are distribution diagrams of the monitoring scenes of pixels corresponding to the target scene at three different times in sequence, and FIGS. 4A, 4B, and 4C are the images of FIGS. 3A, 3B, and 3C The probability distribution of the scene class that the pixels in the region P1 have. Referring first to FIGS. 3A and 4A , in this embodiment, each pixel in the non-visible light dynamic image has the above-mentioned probability distributions of these scene categories (as shown in FIG. 4A ), and the processor 120 is configured to The scene category with the highest probability in the probability distribution of scene categories of the pixel is set as a monitoring scene of the pixel. In this embodiment, the scene categories included in the probability distribution of these scene categories of each pixel include floor (eg scene category A in 4A), bed (eg scene category B), seat (eg scene category C) ) and undefined categories (such as scene category D). In addition, in the present embodiment, the processor 120 is configured to update the state effective block according to the scene of the at least one state effective block A1 of the above at least one image frame and each pixel of the at least one state effective block A1 The probability distribution of the scene class for each pixel in A1.

For example, after the installation of the image monitoring device 100 is completed, the monitoring scene of all pixels of the non-visible light dynamic image of the entire scene is preset as scene category D (ie, the category is undefined), and scene category D is also the default of the pixel. category. At this time, the installer can walk on the floor 52 , while the processor 120 performs operations and judgments on a plurality of image frames, and determines that the status category of the living body 60 (ie, the installer) in each image frame is a station, and determines its status. The corresponding state valid block, and then according to the state type of the living body 60 in each image frame, update the scene type of the pixel of the state valid block (the left and right sides in FIG. 3A ) corresponding to the living body 60 in each image frame. probability distribution. In this embodiment, the state class of the station corresponds to the scene class A (that is, the floor 52 ). Therefore, in the probability distribution of the scene class of the pixels of the valid state block (the left and right sides in FIG. 3A ), the scene class A ( That is, the probability of the floor 52) increases, and exceeds the scene category B, scene category C and scene category D. Therefore, the processor 120 sets the monitoring scene of the pixels in the valid state block (the left and right sides in FIG. 3A ) as scene category A (as shown in FIG. 3A ). In addition, the installer does not have a walking and standing area, such as the center of the target scene, so the area where the installer does not walk and stand (the area adjacent to the center) in Figure 3A does not add and accumulate scene category information, making the probability of the scene category The distribution has not been updated or changed, so the monitoring scene remains the default category, namely scene category D.

Next, the installer lies down in the central area of the target scene, and maintains the lying state for a period of time, and the processor 120 performs operations and judgments on several image frames during this period, and judges the living body 60 ( That is, the state class of the installer) is lying down and its corresponding state valid block is determined, and then according to the state class of the living body 60 in each image frame, the state valid block corresponding to the living body 60 in each image frame is updated (that is, the state valid block corresponding to the living body 60 in each image frame is updated. The probability distribution of the scene class of the pixels in the target scene near the center). In this embodiment, the state category of lying is corresponding to scene category B (that is, the category corresponding to the bed), so the probability distribution of the scene category of the pixels in the valid state block (that is, the area near the center in the target scene) is in the scene category , the probability of scene category B (that is, the category corresponding to the bed) increases. When the probability of scene category B becomes the highest probability in the probability distribution of scene categories, the processor 120 sets the monitoring scene of the pixels of the valid state block (ie, the area near the center in the target scene) as scene category B.

However, at the boundary between the valid state block (that is, the area near the center in the target scene) and the left and right areas (for example, the area P1), the probability distribution of the scene category has not yet been clearly defined, that is, the area P1. In the probability distribution of the scene categories of the pixels within the region, when the probability of scene category A is similar to the probability of scene category B, the processor 120 fails to judge the scene category for the region P1. At this time, the installer can continue to lie down, and can also move his body to change or expand the lying position, and continue to accumulate image frames for the processor 120 to perform operations and judgments. After a certain period of time, as shown in FIG. 3C and FIG. As shown in FIG. 4C , in the probability distribution of the scene categories of the pixels in the area P1, the probability of the scene category B becomes the largest among all the scene categories. At this time, the processor 120 determines that the monitoring scene of all the pixels in the area P1 is the scene category B ( i.e. the category corresponding to bed 54). So far, although the non-visible light dynamic image (such as thermal image) does not include more sensitive details such as human face, clothing, body surface and indoor furnishings, the processor 120 can still determine that the range of the bed 54 is the one shown in FIG. 3C . The range in which the scene category B is located, so as to judge the abnormality accordingly.

In this embodiment, the image monitoring device 100 further includes a storage 130 electrically connected to the processor 120 , wherein the processor 120 is configured to store the non-visible light dynamic image and the scene category corresponding to each pixel in the storage 130 . For example, the processor 120 may store the data of the probability distribution of the scene types as shown in FIG. 3C in the storage 130, and may also store the monitoring scene of each pixel of the non-visible light dynamic image in the storage 130, so as to be regarded as abnormal Basis for Activity Judgment. The storage 130 is, for example, a hard disk, flash memory, random access memory, or other suitable storage. In the above-mentioned embodiment, the monitoring scene of each pixel of the non-visible light dynamic image of the target scene is constructed by the activities of the installers. However, in other embodiments, the activities of the care recipients can also be constructed, or by other people. construct.

In addition, in this embodiment, the processor 120 is configured to perform an operation according to another image frame of the non-visible light dynamic image, and determine the state corresponding to a monitored living body (eg, a person in care) in the target scene to be the above-mentioned states One of the categories and at least one detection valid block corresponding to the monitored living body, according to the at least one detection valid block of the monitored living body, the state of the monitored living body, and the at least one detection valid block of the monitored living body The monitoring scene or scene corresponding to the pixels of the monitored living body is judged whether the state of the monitored living body is abnormal, and a warning signal is output when the state of the monitored living body is judged to be abnormal. The computer or monitoring system in the office, or sending the warning signal to the monitoring host or computer of the remote monitoring center through the Internet.

For example, when the processor 120 determines that the state type of the monitored living body is lying down, and the pixels of the detection effective block of the monitored living body are scene type A (ie, the floor 52 ), and the state of the monitored living body is maintained for a predetermined time (eg, After 30 minutes), the processor 120 determines that the monitored living body (such as the care recipient) has been lying on the floor 52 for too long and has an abnormal condition, so the processor 120 outputs a warning signal to notify the nursing or medical unit personnel to come and check, Or notify the personnel of the remote monitoring center to notify others to come and check. Or, when the processor 120 determines that the state category of the monitored living body (eg, the person being cared for) is lying down, and the pixels of the detection effective block of the monitored living body are scene category B (ie, the bed 54 ), and the state of the monitored living body exceeds another When the preset time (for example, more than 12 hours) is exceeded, the processor 120 determines that the monitored living body state is abnormal, such as being unable to get up due to a physical condition, and outputs a warning signal.

The operation judgment for obtaining the detection valid block in this embodiment is the same as the status valid block in the previous embodiment, wherein the detection valid block in this embodiment is determined according to the status of the monitored living body (such as the person being cared for), and the aforementioned The status valid block of the embodiment is determined according to the status of a living body (eg, an installer).

In one embodiment, the processor 120 is, for example, a central processing unit (CPU), a microprocessor (microprocessor), a digital signal processor (DSP), a programmable controller, a programmable controller, or a programmable controller. A logic device (programmable logic device, PLD) or other similar devices or a combination of these devices is not limited by the present invention. Furthermore, in one embodiment, each function of the processor 120 may be implemented as multiple codes. These code codes are stored in the memory and executed by the processor 120 . Alternatively, in one embodiment, the functions of processor 120 may be implemented as one or more circuits. The present invention does not limit the implementation of the functions of the processor 120 by means of software or hardware.

In addition, in another embodiment, as shown in FIG. 5 , the number of living bodies in the target scene may be multiple, and the number of the corresponding state valid blocks is also multiple. In this embodiment, multiple Two living bodies are taken as an example (the first living body 61 and the second living body 62 are used as examples in FIG. 5 ). The second living body 62 and its corresponding second state valid block B2 are described. The processor 120 is configured to perform the following steps. First, the processor 120 performs an operation according to at least one image frame of the non-visible light dynamic image, and determines that the state of the first living body 61 corresponding to the target scene is one of several state categories, and the state of the second living body 62 is one of the several states. One of the categories, and determines at least one first state valid block B1 where the non-visible light dynamic image corresponds to the state of the first living body 61 , and at least one second state valid block B2 where the non-visible light dynamic image corresponds to the state of the second living body 62 . Next, according to the state type of the first living body 61 , the scene of each pixel of the first state effective block B1 is set as one of several scene types, and each pixel of the second state effective block B2 is set according to the state type of the second living body 62 . The scene of the pixel is one of the several scene categories. For example, the processor 120 determines that the state type of the first living body 61 is lying, and determines its corresponding first state valid block B1, and at the same time, the processor 120 determines that the state type of the second living body 62 is standing, and determines its corresponding The second state of the valid block B2. Next, the processor 120 updates the probability distribution of the scene types of the pixels in the first state valid block B1 according to the state type of the first living body 61 , and at the same time updates the second state valid block B2 according to the state type of the second living body 62 . The probability distribution of the pixel's scene class. In this embodiment, in the probability distribution of the scene category of the pixels in the first state valid block B1, the probability of the scene category B corresponding to the lying state (ie, the category corresponding to the bed) increases; in addition, the second state valid block In the probability distribution of the scene category of the pixels of B2, the probability of scene category A (ie, the floor) corresponding to the state station increases. The processor 120 determines the monitoring scene of each pixel of the non-visible light moving image according to the probability distribution of the scene type of each pixel.

FIG. 6 is a flowchart of an image monitoring method according to an embodiment of the present invention. Referring to FIG. 1 , FIG. 2 and FIG. 6 , the image monitoring method of this embodiment can be implemented by the above-mentioned image monitoring device 100 . The image monitoring method includes the following steps. First, step S210 is executed to obtain a non-visible light dynamic image of the target scene. Next, step S220 is performed to perform an operation according to at least one image frame of the non-visible light dynamic image to determine that the state corresponding to at least one living body 60 in the target scene is one of several state categories and at least one state valid block A1 of the non-visible light dynamic image . Afterwards, step S230 is executed to set a scene of each pixel of the at least one status valid block A1 as one of several scene categories according to the status category of the at least one living body 60 . For some details of the image monitoring method, reference may be made to the matters performed by the image monitoring device 100 described above, which will not be repeated here. The following will describe the matters executed by the above-mentioned image monitoring device 100 and the more detailed steps of the image monitoring method of this embodiment, as shown in FIG. 7 .

Please refer to FIG. 7 , which shows the detailed steps of steps S220 and S230 , which are described as follows. In this embodiment, the non-visible light dynamic image is thermal image data, and after obtaining the non-visible light dynamic image, the processor 120 executes step S104 to perform color gamut conversion on at least one image frame of the non-visible light dynamic image (thermal image frame). image data) is converted from single-channel to three-channel color information. The processor 120 executes step S106, and performs a normalization operation on the output after the color gamut conversion in step S104 to enhance the contrast of different temperatures in the image. For example, normalization may be performed according to the highest temperature in a temperature range to Highlights the contrast of different temperatures within this temperature range in the image. Next, the processor 120 executes step S108, and executes machine learning according to the result of the operation processing in step S106, and calculates the heat source in the non-visible light dynamic image, that is, the state type and area of the living body, which is used to determine the non-visible light dynamic image. the living body.

Step S110 is then executed to perform an operation on the image frame and information of the non-visible light dynamic image obtained by the operation in step S108 to determine the in vivo framed area corresponding to the living body in the image frame of the non-visible light dynamic image, such as the non-visible light in FIG. 8A . In the image frame of the moving image, the living body framing block A2 corresponding to the living body is determined, and an operation is performed to shrink the living body framing block A2 to the living body framing block A3, that is, it is determined that the living body framing block corresponding to the living body is composed of The limbs (living frame area A2 ) are contracted to include the trunk (living frame area A3 ), and the details of step S110 further include steps S112 and S114 . In step S112, the processor calculates the total number of pixels in the Y-axis direction (ie, the vertical axis direction) in the living body framed block A2 one by one along the X-axis direction (ie, the horizontal axis direction) in the living body framed block A2, and the frame is determined by the accumulation of the total number of pixels. After the calculation, the X-axis coordinate corresponding to the maximum number of pixels is extended to the left and right to 30% of the maximum value, which is the block boundary. In step S114, the processor calculates the total number of pixels in the X-axis direction (ie, the horizontal axis direction) in the living body framed block A2 one by one along the Y-axis direction (ie, the vertical axis direction) in the living body framed block A2, and the frame is determined by the cumulative After the operation, the Y-axis coordinate corresponding to the maximum number of pixels is extended up and down to 30% of the maximum value, which is the block boundary. After performing step S110 (including steps S112 and S114 ), the living body framed block A2 converges to the living body framed block A3 , that is, the scope of the living body itself (the body-based block) is determined.

Then, step S116 is executed, the processor 120 performs operations according to the living body framed block A3 and the state type to obtain a state valid block, and the details of step S116 further include steps S118 , steps S120 and steps S122 . In step S118, the processor 120 determines that the state type of the living body is standing or lying down, or in other embodiments, it can also determine that the state type is standing, sitting or lying down. In the case of judging that the state of the living body is standing, step S120 is executed, and the area with a height of 50 pixels below the living body framed block A3 generated after the execution of steps S112 and S114 is extracted as the state effective block A4, and the state effective area is set The scene class of each pixel of block A4 is floor 52, as shown in FIG. 8A. However, the present invention is not limited to a height range of 50 pixels, and in other embodiments, the height of other numbers of pixels may also be used. When it is judged that the state of the living body is lying down, step S122 is executed, and the living body framed block generated after the execution of steps S112 and S114 is set as the state valid block A1, and the scene type is set as bed 54, as shown in FIG. 2 .

After step S120 or step S122 is performed, step S124 is performed to update the probability distribution of the scene type of the pixels in the state valid block, and the details of step S124 further include step S126, step S128, step S130 and step S132. In step S126 , the processor 120 determines whether the pixels in the state-valid block already have scene category information, that is, whether there is a category other than scene category D (ie, undefined category). If the information of the scene type already exists, step S128 is executed, and the processor 120 increases or decreases the probability distribution of the scene type according to the scene type of the pixels in the state valid block. Then, step S130 is executed to determine whether the scene category with the highest probability in the probability distribution of scene categories of each pixel has changed. If the scene category with the highest probability changes, step S132 is executed to update the monitoring scene, for example, the monitoring scene in the area P1 is updated from the scene category of the pixels in the area P1 in FIG. 3B to the scene category of the pixels in the area P1 in FIG. 3C . If the scene category with the highest probability has not changed, step S126 is executed again. In step S126, if it is determined that there is no category definition, step S132 is executed to update the monitoring scene.

It should be noted that, in the embodiments of the present disclosure, the state category includes at least one of standing, sitting, lying down, crawling, and undefined, which is taken as an example to illustrate. , to increase or decrease the status category; in addition, the scene category can also be increased or decreased according to the monitoring environment, monitoring needs or priorities. In some embodiments, the scene category can also be the same as the state category, that is, the scene of the pixel is available for standing, walking or lying down; in another embodiment, the scene category can also include permission or prohibition, that is, setting non-visible light The pixels of the blocks where living bodies (such as installers) appear in the moving image are allowed scene categories, while the pixels of non-visible light moving images (or blocks that have not been updated) are the prohibited scene categories. As a monitoring of theft or security, the present invention is not limited to care.

To sum up, in the image monitoring device and method of the embodiments of the present invention, the non-visible light dynamic image is used to identify the living body, the state type and the state valid block, and the state valid block is set according to the state type of the living body. A pixel's scene is one of several scene categories. Therefore, the image monitoring device and method in the embodiments of the present invention can provide good and effective security monitoring without infringing on the privacy of the care recipient.

52: Flooring 54: Bed 56: Seat 60: Living 61: The first living body 62: Second Living Body 100: Video monitoring device 110: Image Sensing Module 120: Processor 130: Storage A, B, C, D: Scene categories A1: Status valid block A2, A3: In vivo framed block B1: The first state valid block B2: Second state valid block P1: area S102～S132, S210～S230: Steps

FIG. 1 is a schematic diagram of an image monitoring device according to an embodiment of the present invention. FIG. 2 shows a non-visible light dynamic image obtained by the image monitoring device of FIG. 1 . 3A, FIG. 3B, and FIG. 3C are distribution diagrams of monitoring scenes of pixels corresponding to the target scene at three different times in sequence. FIG. 4A , FIG. 4B and FIG. 4C are the probability distributions of scene types that the pixels in the region P1 of FIGS. 3A , 3B and 3C have. FIG. 5 is a schematic diagram of a non-visible light dynamic image obtained by an image monitoring device according to another embodiment of the present invention. FIG. 6 is a flowchart of an image monitoring method according to an embodiment of the present invention. FIG. 7 is a flowchart of the detailed steps of steps S220 and S230 in FIG. 6 . FIG. 8A is a schematic diagram of shrinkage of the living body framed block in steps S110 to S114 in FIG. 7 . FIG. 8B is a schematic diagram showing that the scene type of the 50-pixel height area below the living body framed area in step S120 in FIG. 7 is the floor.

52: Flooring

54: Bed

56: Seat

100: Video monitoring device

110: Image Sensing Module

120: Processor

130: Storage

Claims (22)

An image monitoring device, comprising: an image sensing module configured to obtain a non-visible light dynamic image of a target scene, and the non-visible light dynamic image includes a plurality of image frames; and a processor configured to perform: Perform an operation according to at least one image frame of the non-visible light dynamic image to determine that the state corresponding to at least one living body in the target scene is one of several state categories and at least one state valid block of the non-visible light dynamic image; and According to the state class of the at least one living body, a scene of each pixel of the at least one state valid block is set as one of several scene classes. The image monitoring device according to claim 1, wherein the non-visible light dynamic image is a thermal image, a radio frequency echo image or an ultrasonic image. The image monitoring device of claim 1, wherein the at least one living body is a human body, and the state categories include at least one of standing, sitting, lying, crawling, and undefined. The image monitoring device of claim 3, wherein the processor is configured to further execute: When it is determined that the state type of the at least one living body is a station, setting the scene of each pixel of the at least one state valid block as a floor; When determining that the state type of the at least one living body is sitting, setting the scene of each pixel of the at least one state valid block as a seat; and When it is determined that the state type of the at least one living body is lying down, the scene of each pixel of the at least one state valid block is set as a bed. The image monitoring device of claim 1, wherein each pixel in the non-visible light dynamic image has probability distributions of the scene categories, the processor is configured to divide the probability distributions of the scene categories for each pixel into a probability The highest scene class is set to a monitoring scene for that pixel. The image monitoring device of claim 5, wherein the processor is configured to update the scene according to the at least one state valid block of the at least one image frame and each pixel of the at least one state valid block The probability distribution of the scene categories for each pixel in the state valid block. The image monitoring device of claim 1 or 5, wherein the scene categories include at least one of floor, bed, seat, and undefined categories. The image monitoring device of claim 1, further comprising a storage electrically connected to the processor, wherein the processor is configured to store the non-visible light dynamic image and the scene category corresponding to each pixel in the storage . The image monitoring device of claim 1, wherein the processor is configured to perform an operation according to another image frame of the non-visible light dynamic image to determine the state of a monitored living body corresponding to the target scene as the state categories one and at least one detection valid block corresponding to the monitoring living body, according to the at least one detection valid block of the monitoring living body, the state of the monitoring living body and the at least one detection valid block of the monitoring living body corresponding in the scene, determine whether the state of the monitored living body is abnormal, and output a warning signal when judging that the state of the monitored living body is abnormal. The image monitoring device of claim 5, wherein the processor is configured to perform an operation according to another image frame of the non-visible light dynamic image to determine the states corresponding to a monitored living body in the target scene as the states One of the categories and at least one detection valid block corresponding to the monitored living body, according to the at least one detection valid block of the monitored living body, the state of the monitored living body, and the at least one detection valid area of the monitored living body The monitoring scene corresponding to the block determines whether the state of the monitored living body is abnormal, and outputs a warning signal when judging that the state of the monitored living body is abnormal. The image monitoring device according to claim 1, wherein the at least one living body is a plurality of living bodies, the at least one status valid block is a plurality of status valid blocks, the living bodies correspond to the status valid blocks respectively, and The processor is configured to perform: Perform an operation according to at least one image frame of the non-visible light dynamic image to determine that each of the states corresponding to the living bodies in the target scene is one of several state categories and the state valid blocks of the non-visible light dynamic image; and According to the state class of each of the living bodies, the scene of each pixel of the corresponding state valid block is set as one of several scene classes. An image monitoring method, comprising: obtaining a non-visible light dynamic image of a target scene; Perform an operation according to at least one image frame of the non-visible light dynamic image to determine that the state corresponding to at least one living body in the target scene is one of several state categories and at least one state valid block of the non-visible light dynamic image; and According to the state class of the at least one living body, a scene of each pixel of the at least one state valid block is set as one of several scene classes. The image monitoring method according to claim 12, wherein the non-visible light dynamic image is thermal image, radio frequency echo image or ultrasonic image. The image monitoring method of claim 12, wherein the at least one living body is a human body, and the state categories include at least one of standing, sitting, lying, crawling, and undefined. The image monitoring method according to claim 14, further comprising: When it is determined that the state type of the at least one living body is a station, setting the scene of each pixel of the at least one state valid block as a floor; When determining that the state type of the at least one living body is sitting, setting the scene of each pixel of the at least one state valid block as a seat; and When it is determined that the state type of the at least one living body is lying down, the scene of each pixel of the at least one state valid block is set as a bed. The image monitoring method of claim 12, wherein each pixel in the non-visible light dynamic image has probability distributions of the scene categories, and the image monitoring method further comprises adding the probability distributions of the scene categories of each pixel into the probability distribution of the scene categories. The scene category with the highest probability is set as a monitoring scene of the pixel. The image monitoring method of claim 16, further comprising updating the state valid block according to the at least one state valid block of the at least one image frame and the scene of each pixel of the at least one state valid block The probability distribution of the scene categories for each pixel in . The image monitoring method according to claim 12 or 16, wherein the scene categories include at least one of floor, bed, seat and undefined categories. The image monitoring method according to claim 12, further comprising storing the non-visible light dynamic image and the scene category corresponding to each pixel in a memory. The image monitoring method according to claim 12, further comprising performing an operation according to another image frame of the non-visible light dynamic image to determine that the state of a monitored living body corresponding to the target scene is one of the state categories and corresponding to The at least one detection valid block of the monitored living body is determined according to the at least one detection valid block of the monitored living body, the state of the monitored living body, and the scene corresponding to the at least one detection valid block of the monitored living body. Whether the state of the monitored living body is abnormal, and output a warning signal when judging that the state of the monitored living body is abnormal. The image monitoring method according to claim 16, further comprising performing an operation according to another image frame of the non-visible light dynamic image to determine the state of a monitored living body corresponding to the target scene as one of the state categories and corresponding In at least one detection valid block of the monitored living body, according to the at least one detection valid block of the monitored living body, the state of the monitored living body, and the monitoring scene corresponding to the at least one detection valid block of the monitored living body , judging whether the state of the monitored living body is abnormal, and outputting a warning signal when judging that the state of the monitoring living body is abnormal. The image monitoring method according to claim 12, wherein the at least one living body is a plurality of living bodies, the at least one status valid block is a plurality of status valid blocks, the living bodies correspond to the status valid blocks respectively, and The image monitoring method includes: Perform an operation according to at least one image frame of the non-visible light dynamic image to determine that each of the states corresponding to the living bodies in the target scene is one of several state categories and the state valid blocks of the non-visible light dynamic image; and According to the state class of each of the living bodies, the scene of each pixel of the corresponding state valid block is set as one of several scene classes.

Images