Acting body existence estimation system
The system estimates the presence of action objects covered by sheet-like bodies using feature and positional relationships, addressing the detection challenge and enabling early accident detection with reduced computational load.
Patent Information
- Application Number
- JP2023123663
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-07-28
- Publication Date
- 2025-02-07
AI Technical Summary
Existing behavior estimation systems fail to detect and estimate the presence of an action object when it is covered by a sheet-like body, such as a blanket, due to the joint being obscured and undetectable.
A system that utilizes feature amounts and positional relationships of multiple sites of interest in sample images of action objects covered by flexible and shielding sheet-like bodies, employing machine learning techniques to estimate the presence of the action object by detecting these features and relationships.
Enables reliable estimation of the existence of action objects covered by sheet-like bodies, facilitating early detection of accidents like accidental deaths or falls, and reducing computational load by not requiring identification of specific body parts.
Smart Images
Figure 2025019825000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to a moving object presence estimation system capable of estimating the presence of a moving object wearing a sheet-like material. [Background technology]
[0002] 2. Description of the Related Art Conventionally, there has been known a behavior estimation device that estimates the behavior of a moving object based on the coordinates of a plurality of joints of the moving object and the displacement of depth in a plurality of time-series images (see, for example, Patent Document 1). [Prior art documents] [Patent documents]
[0003] [Patent Document 1] Patent No. 6525179 Summary of the Invention [Problem to be solved by the invention]
[0004] In the above technology, the presence of a moving object is estimated by detecting joints, but in the case of a moving object wrapped in a sheet-like object such as a blanket, the joints cannot be detected and the presence of the moving object cannot be estimated.
[0005] Therefore, an object of the present invention is to provide a moving object presence estimation system capable of estimating the presence of a moving object wrapped in a sheet-like material. [Means for solving the problem]
[0006] The present invention provides a moving object presence estimation system comprising: a memory unit that stores feature values and positional relationships of multiple target parts of a sample moving object that appear in the shape of a sheet-like material having flexibility and shielding properties in a sample image in which the sample moving object is wrapped around the sheet-like material; an estimated side acquisition unit that acquires a target image; an estimated side detection unit that detects multiple feature values in the target image; and an estimation unit that estimates that a target moving object wrapped around a sheet-like material having flexibility and shielding properties is present in the target image if the detected multiple feature values include those that correspond to the feature values and positional relationships of the multiple target parts.
[0007] According to another aspect of the present invention, there are provided a moving object presence estimation program and a moving object presence estimation method corresponding to the moving object presence estimation system.
[0008] According to another aspect of the present invention, a learning device is provided which includes a learning side acquisition unit which acquires a sample image in which a sample moving object is wrapped around a sheet-like body having flexibility and shielding properties, a position information input unit to which position information of multiple target parts of the sample moving object appearing in the shape of the sheet-like body in the sample image is input, a learning side detection unit which detects multiple feature amounts in the sample image, and a learning unit which learns the positional relationship of the feature amounts of the detected feature amounts of the part for which the position information was input.
[0009] According to another aspect of the present invention, there is provided a learning program and a learning method corresponding to the above learning device. Effect of the Invention
[0010] According to the moving object presence estimation system of the present invention, it is possible to estimate the presence of a moving object wearing a sheet-like material. [Brief description of the drawings]
[0011] [Figure 1] FIG. 1 is an explanatory diagram of an image used in a moving object presence estimation system according to a first embodiment of the present invention; [Diagram 2] 1 is a block diagram of a moving object presence estimation system according to a first embodiment of the present invention; [Diagram 3] FIG. 1 is an explanatory diagram of position information of a region of interest according to a first embodiment of the present invention; [Figure 4] 1 is a flow chart of a moving object presence estimation system according to a first embodiment of the present invention. [Diagram 5] FIG. 11 is an explanatory diagram of an image used in the moving object presence estimation system according to the second embodiment of the present invention; [Figure 6] FIG. 13 is an explanatory diagram of an image used in the moving object presence estimation system according to the third embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0012] Hereinafter, a moving object presence presumption system 1 according to a first embodiment of the present invention will be described with reference to FIGS. 1 to 4. FIG.
[0013] 1, the moving object presence estimation system 1 is for estimating the presence of a moving object Y wearing a sheet-like object S within a target image X. FIG 1 shows an example of a person sitting cross-legged wearing the sheet-like object S.
[0014] The sheet-like body S has flexibility such that when the active object Y wraps around it, a plurality of target areas of the active object Y are reflected in the shape of the sheet-like body S, and may be, for example, a cloth (including a comforter, a blanket, etc.), a vinyl sheet, etc. The target areas may be, for example, the head, shoulders, waist, knees, etc.
[0015] In this embodiment, "appears in the shape of the sheet-like body S" does not necessarily mean that the shape of the focused area is clearly shown. For example, when wrapped in thin cloth, the shape of the focused area is expected to be clearly shown, but when wrapped in a thick blanket, the shape of the focused area is expected to be indistinguishable. Therefore, in this embodiment, "appears in the shape of the sheet-like body S" means, in other words, that "the shape of the sheet-like body S is deformed due to the focused area."
[0016] Furthermore, the sheet-like body S has a shielding property such that the active object Y is at least partially shielded when wrapped around the sheet-like body S. Note that the shielding property in this embodiment does not necessarily mean that the active object Y is completely shielded, but also includes a case where the active object Y is shielded to an extent that it is difficult to detect by the learning side detection unit 23 and the estimation side detection unit 33 described later.
[0017] Furthermore, with regard to "wrapping" in this embodiment, it does not necessarily have to be completely wrapped around, but also includes cases where it wraps around to the extent that necessary parts of the action object Y (joints, etc.) are difficult to detect by the learning side detection unit 23 and the estimation side detection unit 33 described later.
[0018] The active object presence estimation system 1 includes a learning device 2 and an estimation device 3, as shown in FIG.
[0019] First, the learning device 2 will be described.
[0020] As shown in FIG. 2, the learning device 2 includes a learning-side acquisition unit 21, a position information input unit 22, a learning-side detection unit 23, and a learning unit 24.
[0021] The learning side acquiring unit 21 acquires a sample image X1 in which a sample moving object Y1 wearing a sheet-like body S1 as shown in FIG.
[0022] In this embodiment, it is assumed that a user of the active object presence estimation system 1 inputs a sample image X1 captured by a photographing means to the learning side acquisition unit 21, so that the learning side acquisition unit 21 acquires the sample image X1.
[0023] The position information input unit 22 receives position information Z of a plurality of target regions of a sample moving object Y1 appearing in the shape of the sheet-like body S1 in the sample image X1.
[0024] The part of interest may be, for example, the head, shoulders, waist, knees, etc., and information indicating the position of the part of interest is input as position information Z to the corresponding part of the sample image X1 by the user of the moving object presence estimation system 1 (for example, as shown in FIG. 3, an annotation indicating the part of interest is made at the part corresponding to the part of interest). Note that the part of interest is not limited to the head, shoulders, waist, and knees, and the user may input the part of interest that appears on the shape of the sheet-like body S.
[0025] The learning-side detection unit 23 detects a plurality of feature amounts in the sample image X1. For example, a method of extracting feature amounts for each set of a predetermined number of pixels (N×M pixels) in the sample image X1 using a convolution operation can be considered.
[0026] In this embodiment, the learning-side detection unit 23 only needs to detect the features of the entire sample image X1 (or a specified part, if necessary), and does not need to specify "which part of the body the detected features correspond to."
[0027] The learning unit 24 learns the positional relationship of the feature amounts of the portion to which the position information Z is input, out of the detected feature amounts.
[0028] Here, different feature amounts are detected at the target parts (head, shoulders, waist, knees, etc.) of the sample moving object Y1 even through the sheet-like body S1. Therefore, the learning unit 24 learns the positional relationship of the feature amounts of the target parts (head, shoulders, waist, knees, etc.) of the sample moving object Y1 through the sheet-like body S1.
[0029] For example, as shown in FIG. 3(a), in the case of a sample image X1 in which a sample subject Y1 in a cross-legged position is wearing a sheet-like body S1, it is expected that the parts corresponding to the head, shoulders, and knees will be mainly represented in the shape of the sheet-like body S1, and therefore the positional relationships of these parts will be learned.
[0030] Furthermore, as shown in FIG. 3(b), in the case of sample image X1 in which a sample object Y1 in a reclining position is wrapped in a sheet-like body S1, it is considered that the parts corresponding to the head, shoulders, and waist are mainly represented in the shape of the sheet-like body S1, and therefore the positional relationship of these parts is learned.
[0031] The operations of the learning side detection unit 23 and the learning unit 24 can be performed by machine learning such as deep learning.
[0032] In addition, the learning may include the posture (sitting cross-legged, lying down, etc.), or may only include learning the positional relationship of the feature amount of the part where the position information Z is input. When learning includes the posture, the user may also input the posture information from the position information input unit 22, etc.
[0033] In this way, the learning device 2 learns the feature quantities and positional relationships of the target portion of the sample moving object Y1 using a large number of sample images X1. In learning, it is preferable to use sheet-like bodies S1 of various thicknesses, sizes, materials, etc.
[0034] Next, the estimation device 3 will be described.
[0035] As shown in FIG. 2, the estimation device 3 includes a storage unit 31, an estimation-side acquisition unit 32, an estimation-side detection unit 33, an estimation unit , and a notification unit .
[0036] The storage unit 31 stores feature amounts and positional relationships of multiple target parts of the sample moving object Y1 that appear in the shape of the sheet-like body S1 in the sample image X1 in which the sample moving object Y1 wrapped in the flexible and shielding sheet-like body S1 is captured. In this embodiment, the storage unit 31 stores the feature amounts and positional relationships learned by the learning unit 24.
[0037] The estimation side acquisition unit 32 acquires the target image X2.
[0038] In this embodiment, a plurality of target images X2 in time series are acquired, and the plurality of target images X2 in time series captured by an installed imaging means are automatically input to the estimation side acquisition unit 32.
[0039] The estimation side detection unit 33 detects a plurality of feature amounts in the target image X2. For example, a method of extracting feature amounts for each group of a predetermined number of pixels (N×M pixels) in the target image X2 using a convolution operation can be considered.
[0040] In this embodiment, the estimated side detection unit 33 further continuously or intermittently detects a predetermined part of the target action object Y2 captured in a plurality of target images X2 in a time series. As the predetermined part, for example, it is considered to detect the head of the target action object Y2.
[0041] It is possible to use a known method for detecting the predetermined part, for example, by storing a feature quantity indicating the head in the storage unit 31 or the like in advance, and detecting a feature quantity that is closer to the feature quantity indicating the head than a predetermined value as the head.
[0042] In addition, the estimated side detection unit 33 detects the head of the target behavior object Y2 when the target behavior object Y2 is not wrapped in the sheet-like body S2, but is unable to detect the head of the target behavior object Y2 when the target behavior object Y2 has the sheet-like body S2 wrapped around its head.
[0043] When the detected multiple feature points include those that correspond to the features and positional relationships of multiple target areas, the estimation unit 34 estimates that a target action object Y2 wrapped in a sheet-like material S2 having flexibility and shielding properties is present in the target image X2.
[0044] For example, when the position information Z shown in Figure 3 is detected in a positional relationship (such as a roughly proportional positional relationship) in which a feature value that is closer to the feature value of the inputted area of interest than a predetermined value is detected, it is estimated that a target object Y2 wrapped in a sheet-like body S2 having flexibility and shielding properties is present in the target image X2.
[0045] In this case, if posture information has also been learned in the learning device 2, it is possible to estimate the posture as well, making it possible to estimate the posture in which the target action object Y2 is wrapped around the sheet-like object S2.
[0046] Furthermore, in this embodiment, when a specific part that was detected (in this embodiment, the head) is not detected and the detected multiple features include those that correspond to the features and positional relationships of multiple parts of interest, it is estimated that a target action object Y2 wrapped in a sheet-like body S2 that has flexibility and shielding properties is present in the target image X2.
[0047] In this case, since the specific part that had been detected until then is no longer detected, it becomes possible to more reliably and quickly infer that the entire body of the target behavior object Y2 has been hidden by the sheet-like object S2, such as a futon.
[0048] The notification unit 35 issues a notification when the estimation unit 34 estimates that a target action object Y2 wrapped in a flexible and shielding sheet-like body S2 is present in the target image X2. For example, it is possible to issue a notification to a manager of a facility.
[0049] The above-described active object presence estimation system 1 can be used for early detection of accidental death, suicide, etc., in a state where the person is covered with a blanket or the like.
[0050] Next, the operation of the moving object presence estimation system 1 will be described with reference to the flowchart of Fig. 4. The moving object presence estimation system 1 performs operations in two stages: a learning stage and an estimation stage.
[0051] In the learning stage, a sample image X1 is acquired by the learning side acquisition unit 21 (S1), position information is input to the position information input unit 22 (S2), and features in the sample image X1 are detected by the learning side detection unit 23 (S3). The learning unit 24 then learns the positional relationship of the features of the part of the detected features for which position information Z was input (S4).
[0052] Steps S1-S4 are preferably performed for multiple sample images X1 and sample action objects Y1. Steps S2 and S3 may be performed in reverse order.
[0053] Next, in the estimation stage, the estimation side acquisition unit 32 acquires a target image X2 (S5), and when the estimation side detection unit 33 detects a plurality of feature amounts in the target image X2 (S6), if the detected plurality of feature amounts include those corresponding to the feature amounts and positional relationships of a plurality of focused parts (S7: YES), the estimation unit 34 estimates that a target action object Y2 wrapped in a flexible and shielding sheet-like body S2 is present in the target image X2 (S8). Thereafter, the notification unit 35 issues a notification as necessary.
[0054] As described above, in the moving object presence estimation system 1 of this embodiment, if the multiple feature amounts detected in the target image X2 include those corresponding to the feature amounts and positional relationships of multiple target areas, it is estimated that a target moving object Y2 wrapped around a sheet-like object S having flexibility and shielding properties is present in the target image X2.
[0055] With this configuration, even if the target behavior entity Y2 is wrapped in a sheet-like body S2 with shielding properties, it is possible to estimate the presence of the target behavior entity Y2 in the target image X2 by referring to the feature amounts and positional relationships of multiple target parts of the sample behavior entity Y1 that appear in the shape of the sheet-like body S1, making it possible to discover accidental deaths, suicides, etc. while covered with a blanket, etc. Furthermore, in both the learning device 2 and the estimation device 3, there is no need to estimate "whether a behavior entity exists" (to identify parts, etc. of the sample behavior entity Y1 and target behavior entity Y2), making it possible to significantly reduce the load on the computer.
[0056] In addition, in the moving object presence estimation system 1 of this embodiment, when a specific part (such as the head) that was detected is not detected and the detected multiple features include those that correspond to the features and positional relationships of multiple parts of interest, it is estimated that a target moving object Y2 wrapped around a sheet-like body S2 that is flexible and has shielding properties is present in the target image X2.
[0057] With this configuration, since a specific part that had been detected until then is no longer detected, it becomes possible to more reliably and quickly infer that the entire body of the target behavior object Y2 is hidden by a sheet-like object S2 such as a futon.
[0058] Next, a moving object presence presumption system 10 according to a second embodiment of the present invention will be described with reference to FIGS.
[0059] In this embodiment, attention is focused on the fact that when a person falls from a bed, the person falls from the waist or buttocks, and it is estimated that there is a risk that the target action object Y2 wearing the sheet-like body S will fall from the bed.
[0060] The configuration of the moving object presence estimation system 10 according to this embodiment is the same as the configuration of the moving object presence estimation system 1 according to the first embodiment shown in FIG. 2, so with regard to FIG. 2, the same as in the first embodiment will be referred to.
[0061] On the other hand, in this embodiment, the memory unit 31 further stores features of at least one of the waist and buttocks of the sample moving object Y1 that are reflected in the shape of the sheet-like body S1 in the sample image X1 showing the sample moving object Y1 lying down and wrapped in the sheet-like body S1.
[0062] As in the first embodiment, the feature quantities of at least one of the waist and buttocks of the sample moving body Y1 appearing in the shape of the sheet-like body S1 can be obtained by inputting position information Z of at least one of the waist and buttocks of the sample moving body Y1 appearing in the shape of the sheet-like body S1 in the sample image X1 from the position information input unit 22, and learning the positional relationship of the feature quantities of the part of the feature quantities detected by the learning side detection unit 23 for which the position information Z was input.
[0063] In this embodiment, the estimated side detection unit 33 can further detect the position of an end E of the bed reflected in the target image X2, as shown in FIG.
[0064] A known method may be used to detect the position of the end E of the bed. For example, a method may be considered in which a feature amount indicating the bed shape is stored in advance in the storage unit 31 or the like, and a feature amount that is closer than a predetermined value to the feature amount indicating the bed shape is detected as the bed (end E of the bed). Note that, although Fig. 5 shows an example in which the bed is photographed from above, the position of the end E of the bed may be detected in the same manner even if the bed is photographed from another direction, such as diagonally above.
[0065] Then, the estimation unit 34 estimates that there is a risk of the target action object Y2 falling from the bed when the detected multiple feature amounts include those corresponding to the feature amounts and positional relationships of multiple target areas, and those corresponding to the feature amounts of at least one of the waist and buttocks, and when the one corresponding to the feature amounts of at least one of the waist and buttocks is located within a range R set for the end E of the bed.
[0066] Here, as shown in Fig. 5, at least one of the features of the lower back and buttocks may be located outside the bed, and in such a case, the risk of falling off the bed becomes extremely high. Therefore, for example, as shown in Fig. 5, a method of setting a range R within a predetermined distance close to the end E of the bed as the inside of the bed and setting the entire range as the outside of the bed as the range R can be considered.
[0067] As described above, in the moving object presence estimation system 10 of this embodiment, when something corresponding to the characteristic parameters of at least one of the lower back and buttocks is located within a range R set relative to the end E of the bed, it is estimated that there is a risk of the target moving object Y2 falling from the bed.
[0068] With this configuration, even if the target moving object Y2 is lying down and wrapped in a sheet-like body S2 with shielding properties, it is possible to estimate the position of the waist or buttocks, which may cause the target moving object Y2 to fall off the bed, thereby making it possible to prevent the target moving object Y2 from falling off the bed.
[0069] Next, a moving object presence presumption system 20 according to a third embodiment of the present invention will be described with reference to FIGS.
[0070] In this embodiment, it is estimated that there is a risk of suffocation when sleeping on one's stomach.
[0071] The configuration of the moving object presence estimation system 20 according to this embodiment is the same as the configuration of the moving object presence estimation system 1 according to the first embodiment shown in FIG. 2, so with regard to FIG. 2, the same as in the first embodiment will be referred to.
[0072] On the other hand, in this embodiment, the memory unit 31 further stores the feature amounts and positional relationships of multiple focal areas of the sample moving object Y1 that appear in the shape of the sheet-like body in the sample image X1 in which the sample moving object Y1 is shown lying down and wrapped around the sheet-like body S1.
[0073] In addition, in this embodiment, the estimated side detection unit 33 can further detect the direction of the face of the target actor Y2 captured in the target image X2.
[0074] In this embodiment, the face is not covered with the sheet-like body S2. A known method may be used to detect the direction of the face. For example, the direction of the face may be detected based on the presence or absence of hair. Although FIG. 6 shows an example in which the bed is photographed from the side, the direction of the face may be detected in the same manner even if the bed is photographed from another direction, such as from above.
[0075] Then, the estimation unit 34 estimates that there is a risk of choking when the detected multiple feature amounts include those corresponding to the feature amounts and positional relationships of multiple areas of interest, and when the angle between the detected face direction and the vertically downward is less than a predetermined value.
[0076] For example, as shown in FIG. 6, when the direction of the face is substantially vertically downward, it is considered that the target action object Y2 is in danger of choking, and the notification unit 35 issues a notification.
[0077] As described above, the active object presence presumption system 20 according to this embodiment presumes that there is a risk of choking when the angle between the detected face direction and the vertically downward direction is equal to or smaller than a predetermined value.
[0078] With this configuration, it can be estimated that the target behavior object Y2 is wrapped in a sheet-like body S with shielding properties while lying down, and by further detecting the direction of its face, it is possible to quickly prevent the target behavior object Y2 from suffocating.
[0079] The moving object presence estimation system according to the present invention is not limited to the above-described embodiment, and various modifications and improvements are possible within the scope of the claims.
[0080] For example, in the first embodiment, a specified part (head) is detected, but it is not necessary to detect the specified part (head). It is possible to estimate that a target action object Y2 wrapped in a flexible and shielding sheet-like body S is present in the target image X2 simply by determining whether or not the multiple features detected in the target image X2 include any that correspond to the features of multiple target parts.
[0081] In the second embodiment, it is estimated that the target action object Y2 is in danger of falling off the bed when "something corresponding to the characteristic amount of at least one of the lower back and buttocks is located within a range R set with respect to the end E of the bed", but it may be estimated that the target action object Y2 is in danger of falling off the bed when "something corresponding to the characteristic amount of at least one of the lower back and buttocks is gradually moving toward the outside of the bed". "Moving gradually" may be, for example, when the target action object Y2 is moving from the inside of the bed to the outside by a predetermined distance within a predetermined time.
[0082] In the third embodiment, the sheet-like body S2 is not wrapped around the face, but it may be wrapped around the face.
[0083] In this case, the storage unit 31 further stores the head feature of the sample moving object Y1 that appears in the shape of the sheet-like object S1 in the sample image X1 in which the sample moving object Y1 is lying down and wrapped in the sheet-like object S1, and the estimation side detection unit 33 can detect the facial orientation of the target moving object Y2 that appears in the target image X2 by referring to the head feature of the sample moving object Y1, thereby making it possible to estimate that there is a risk of suffocation even if the face is wrapped in the sheet-like object S2. When detecting the facial orientation of the target moving object Y2 that appears in the target image X2, it is possible to store the head feature of the sample moving object Y1 that appears in the shape of the sheet-like object, taking into account the facial orientation.
[0084] In addition, in the above embodiment, the memory unit 31 stores "the feature values and positional relationships of multiple target parts of the sample moving body that appear in the shape of a sheet-like body wrapped around a flexible and shielding sheet-like body in a sample image showing the sample moving body wrapped around the flexible and shielding sheet-like body" learned by the learning device 2, but it may also store, for example, "the feature values and positional relationships of multiple target parts of the sample moving body that appear in the shape of the sheet-like body in a sample image showing the sample moving body wrapped around a flexible and shielding sheet-like body" set on a rule basis.
[0085] In addition, in the above embodiment, a human being is used as an example of a moving body, but the present invention can also be used for animals and robots.
[0086] Furthermore, the learning device 2 and the estimation device 3 in the active object presence estimation system 1 may be separate entities and be capable of communicating with each other.
[0087] The present invention can also be applied to a program or method corresponding to the processes performed by the learning device 2 and the estimation device 3, and a recording medium storing the program. In the case of a recording medium, the program is installed in a computer or the like. Here, the recording medium storing the program may be a non-transient recording medium. A CD-ROM or the like is considered as a non-transient recording medium, but is not limited thereto. [Explanation of symbols]
[0088] 1, 10, 20 Behavior Presence Estimation System 2 Learning device 3 Estimation device 21 Learning side acquisition unit 22 Location information input section 23 Learning side detector 24 Learning Department 31 Storage section 32 Estimation side acquisition part 33 Estimation side detection unit 34 Estimation part 35 Notification Department E Bed End R Range S, S1, S2 Sheet-shaped body X Image X1 Sample Image X2 Target image Y Action Body Y1 Sample Actions Y2 Target Action Z location information
Claims
1. A storage unit that stores feature amounts and positional relationships of a plurality of target portions of a sample moving body that are shown in the shape of a sheet-like body having flexibility and shielding properties in a sample image showing the sample moving body wrapped around the sheet-like body; An estimation side acquisition unit that acquires a target image; an estimation side detection unit that detects a plurality of feature amounts in the target image; an estimation unit that estimates that a target action object wrapped in a flexible and occluding sheet-like material is present in the target image when the detected multiple feature amounts include those corresponding to the feature amounts and positional relationships of the multiple target areas; A moving object presence estimation system comprising:
2. The estimation side acquisition unit acquires a plurality of target images in a time series, The estimation side detection unit further detects a predetermined part of a target action object captured in the plurality of target images in the time series continuously or intermittently; The moving object presence estimation system described in claim 1, characterized in that the estimation unit estimates that the target moving object wrapped in a flexible, sheet-like material with shielding properties is present in the target image when the specified part that was detected is not detected and the detected multiple feature amounts include those corresponding to the feature amounts and positional relationships of the multiple parts of interest.
3. The storage unit further stores at least one feature of the waist and buttocks of the sample behavioral body, which is represented in the shape of the sheet-like body having flexibility and shielding properties in a sample image showing the sample behavioral body wrapped in the sheet-like body in a recumbent state, The estimated side detection unit is further capable of detecting a position of an edge of a bed shown in the target image, The moving object presence estimation system described in claim 1, characterized in that the estimation unit estimates that the target moving object is at risk of falling from the bed when the detected multiple feature amounts include those corresponding to the feature amounts and positional relationships of the multiple target areas, and those corresponding to the feature amounts of at least one of the waist and buttocks, and when the one corresponding to the feature amounts of at least one of the waist and buttocks is located within a range set relative to the edge of the bed.
4. The storage unit further stores a feature amount of at least one of the waist and buttocks of the sample behavior body that appears in the shape of the sheet-like body in a sample image showing the sample behavior body lying down and wrapped in the sheet-like body, The estimation side acquisition unit acquires a plurality of target images in a time series, The estimated side detection unit is further capable of detecting a position of an edge of a bed reflected in the target image, The moving object presence estimation system of claim 1, characterized in that the estimation unit estimates that the target moving object is at risk of falling from the bed when the detected multiple feature amounts include those corresponding to the feature amounts and positional relationships of the multiple target areas, and those corresponding to the feature amounts of at least one of the waist and buttocks, and when the one corresponding to the feature amounts of at least one of the waist and buttocks is gradually moving toward the outside of the bed.
5. The storage unit further stores feature amounts and positional relationships of a plurality of target portions of the sample behavior body that appear in the shape of the sheet-like body in a sample image showing the sample behavior body lying down and wrapped around the sheet-like body, The estimation side detection unit is further capable of detecting a face direction of a target action object captured in the target image, The moving object presence estimation system described in claim 1, characterized in that the estimation unit estimates that the target moving object is at risk of suffocation when the detected multiple features include those corresponding to the features and positional relationships of the multiple target areas and when the angle between the detected face direction and a vertically downward direction is less than a predetermined value.
6. The storage unit further stores a feature amount of the head of the sample behavior object, which is represented by a shape of the sheet-like body having flexibility and shielding properties in a sample image showing the sample behavior object wrapped around the sheet-like body in a recumbent state, the estimation side detection unit is capable of detecting a facial orientation of a target action object captured in the target image by referring to a feature amount of a head of the sample action object; The moving object presence estimation system described in claim 1, characterized in that the estimation unit estimates that the target moving object is at risk of suffocation when the detected multiple features include those corresponding to the features and positional relationships of the multiple target areas and when the angle between the detected face direction and a vertically downward direction is less than a predetermined value.
7. A program executed by a computer that stores feature amounts and positional relationships of a plurality of target portions of a sample behavior object that are shown in a shape of a sheet-like body having flexibility and shielding properties in a sample image of the sample behavior object, the feature amounts and positional relationships being executed by a computer that stores feature amounts and positional relationships of a plurality of target portions of the sample behavior object that are shown in a shape of the sheet-like body in a sample image of the sample behavior object that is wrapped around the sheet-like body having flexibility and shielding properties, acquiring an image of a target; detecting a plurality of feature amounts in the target image; When the detected plurality of feature amounts includes features corresponding to the plurality of target portions and their positional relationships, it is estimated that a target action object wrapped in a flexible and opaque sheet-like material is present in the target image; A behavioral object presence estimation program comprising:
8. A method executed by a computer that stores feature amounts and positional relationships of a plurality of target portions of a sample behavior object that are shown in a shape of a sheet-like body having flexibility and shielding properties in a sample image of the sample behavior object, the feature amounts and positional relationships of a plurality of target portions of the sample behavior object that are shown in a shape of the sheet-like body in a sample image of the sample behavior object, the method comprising: acquiring an image of a target; detecting a plurality of feature amounts in the target image; When the detected plurality of feature amounts includes features corresponding to the plurality of target portions and their positional relationships, it is estimated that a target action object wrapped in a flexible and opaque sheet-like material is present in the target image; A method for estimating the presence of a moving object, comprising:
9. A learning side acquisition unit that acquires a sample image of a sample moving object wrapped in a sheet-like body having flexibility and shielding properties; a position information input section for inputting position information of a plurality of target portions of a sample action object appearing on the shape of the sheet-like body in the sample image; A learning-side detection unit that detects a plurality of feature amounts in the sample image; a learning unit configured to learn a positional relationship of the feature amount of the portion of the feature amount detected, the portion corresponding to the input position information; A learning device comprising:
10. A program executed on a computer, A step of acquiring a sample image of a sample moving object wrapped in a sheet-like material having flexibility and shielding properties; A step of inputting position information of a plurality of target portions of a sample action object appearing on the shape of the sheet-like body in the sample image; detecting a plurality of features in the sample image; learning a positional relationship of the feature amounts of the portion of the feature amount detected for which the position information is input; A learning program characterized by comprising:
11. 1. A computer-implemented method comprising: A step of acquiring a sample image of a sample moving object wrapped in a sheet-like material having flexibility and shielding properties; A step of inputting position information of a plurality of target portions of a sample action object appearing in the shape of a sheet-like body in the sample image; detecting a plurality of features in the sample image; learning a positional relationship of the feature amounts of the portion of the feature amount detected for which the position information is input; A learning method comprising:
Citation Information
Patent Citations
Action estimation device
JP6525179B1