CN110378244A - The detection method and device of abnormal posture - Google Patents
The detection method and device of abnormal posture Download PDFInfo
- Publication number
- CN110378244A CN110378244A CN201910562307.3A CN201910562307A CN110378244A CN 110378244 A CN110378244 A CN 110378244A CN 201910562307 A CN201910562307 A CN 201910562307A CN 110378244 A CN110378244 A CN 110378244A
- Authority
- CN
- China
- Prior art keywords
- posture
- target user
- key point
- target
- abnormal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
Abstract
This disclosure relates to a kind of detection method and device of exception posture, it is related to intelligent attitude detection field, this method comprises: obtaining the target image of target user, obtain the position coordinates of the key point of target user in the target image, key point includes the human body of pre-set target user, determines whether the posture of target user is abnormal posture according to position coordinates.Whether the posture that the disclosure when user does not have apparent change in displacement, can accurately determine user is abnormal posture, and detection efficiency is high, and does not need user and additionally carry detection device, is easy to use, improves user experience.
Description
Technical field
This disclosure relates to intelligent attitude detection field, and in particular, to a kind of detection method and device of exception posture.
Background technique
With the development of economy with the raising of medical level, average human life constantly extends, social senilization's process
Aggravation.Since self-care ability and self-protection ability decline accidental falls can usually occur during action for old man.Work as old man
When being in alone, in case of accidental falls, it is possible that situations such as brain injury or ability to act are lost, leads to miss and controls
Treatment opportunity affects the state of an illness even threat to life adversely.In order to ensure the safety of old solitary people, need whether real-time detection old solitary people falls
?.In the prior art, the change in displacement of moment is fallen down, by detecting old man using the detection device of contact mainly to judge
Whether old man falls down, but the detection device of contact falls down moment there is no when apparent change in displacement in old man, is difficult to detect
It is fallen down to old man, accuracy and efficiency are lower, also, the detection device of contact needs old man's body-worn and timing is charged,
The comfort level that will affect old man's daily life is unfavorable for promoting.
Summary of the invention
Purpose of this disclosure is to provide the detection methods and device of a kind of abnormal posture, abnormal in the prior art to solve
The accuracy and low efficiency of attitude detection, and need user's body-worn and timing charge the problem of.
To achieve the goals above, according to the first aspect of the embodiments of the present disclosure, a kind of detection side of abnormal posture is provided
Method, this method comprises:
Obtain the target image of target user;
Position coordinates of the key point of the target user in the target image are obtained, the key point includes preparatory
The human body for the target user being arranged;
Whether the posture that the target user is determined according to the position coordinates is abnormal posture.
Optionally, whether the posture that the target user is determined according to the position coordinates is that abnormal posture includes:
The first image-region is determined from the target image according to preset dividing line;
According to the position coordinates and the first image region, determine whether the posture of the target user is abnormal appearance
State.
Optionally, described according to the position coordinates and the first image region, determine the posture of the target user
It whether is that abnormal posture includes:
First object key point is determined from the key point, the first object key point includes the position coordinates position
Key point in the first image region;
Determine the first quantity of the first object key point;
If first quantity is more than or equal to the first preset quantity threshold value, determine that the posture of the target user is different
Normal posture;
If first quantity is less than the first preset quantity threshold value, determine that the posture of the target user is non-exception
Posture.
Optionally, the key point includes at least two key points, described according to position coordinates determination
Whether the posture of target user is that abnormal posture includes:
The key point distance between key point described in every two is determined according to the position coordinates;
Determine whether the posture of the target user is abnormal posture according to the key point distance.
Optionally, described to determine whether the posture of the target user is abnormal posture packet according to the key point distance
It includes:
Determine target range from key point distance, the target range be meet the key point of preset condition away from
From;
Determine the second quantity of the target range;
If second quantity is more than or equal to the second preset quantity threshold value, determine that the posture of the target user is different
Normal posture;
If second quantity is less than the second preset quantity threshold value, determine that the posture of the target user is non-exception
Posture.
Optionally, described to determine whether the posture of the target user is abnormal posture packet according to the key point distance
It includes:
Obtain the corresponding distance weighting of key point described in every two;
Determine target range from key point distance, the target range be meet the key point of preset condition away from
From;
Posture is calculated according to the target range and the distance weighting and determines parameter;
If the posture determines that parameter is more than or equal to parameter preset threshold value, determine that the posture of the target user is different
Normal posture;
If the posture determines that parameter is less than parameter preset threshold value, determine that the posture of the target user is non-abnormal appearance
State.
Optionally, the key point includes multiple key points, described to determine the target according to the position coordinates
Whether the posture of user is that abnormal posture includes:
Determine that the second target critical point, the second target critical point are any in the key point from the key point
Key point;
Reference frame is established centered on the second target critical point;
Other key points in the key point in addition to the second target critical point are mapped to the reference frame
In, obtain the relative position coordinates of other key points;
Using the relative position coordinates as the input of preset convolutional neural networks, to obtain the convolutional neural networks
The first probability and the second probability of output, first probability is that the posture of the target user is abnormal appearance probability of state, institute
Stating the posture that the second probability is the target user is non-abnormal appearance probability of state;
If first probability is greater than or equal to second probability, determine the posture of the target user for abnormal appearance
State;
If first probability is less than second probability, determine that the posture of the target user is non-abnormal posture.
Optionally, the method also includes:
If it is determined that the posture of the target user is abnormal posture, the first warning information is issued;And/or
If it is determined that the posture of the target user is abnormal posture, the second warning information is sent to preset terminal.
According to the second aspect of an embodiment of the present disclosure, a kind of detection device of abnormal posture is provided, described device includes:
Module is obtained, for obtaining the target image of target user;
The acquisition module is also used to obtain position of the key point of the target user in the target image and sits
Mark, the key point includes the human body of the pre-set target user;
Determining module, for determining whether the posture of the target user is abnormal posture according to the position coordinates.
Optionally, the determining module includes:
First determines submodule, for determining the first image-region from the target image according to preset dividing line;
Second determines submodule, for determining that the target is used according to the position coordinates and the first image region
Whether the posture at family is abnormal posture.
Optionally, described second determine that submodule is used for:
First object key point is determined from the key point, the first object key point includes the position coordinates position
Key point in the first image region;
Determine the first quantity of the first object key point;
If first quantity is more than or equal to the first preset quantity threshold value, determine that the posture of the target user is different
Normal posture;
If first quantity is less than the first preset quantity threshold value, determine that the posture of the target user is non-exception
Posture.
Optionally, the key point includes at least two key points, and the determining module includes:
Third determines submodule, for according to the position coordinates determine the key point between key point described in every two away from
From;
4th determines submodule, for determining whether the posture of the target user is abnormal according to the key point distance
Posture.
Optionally, the described 4th determine that submodule is used for:
Determine target range from key point distance, the target range be meet the key point of preset condition away from
From;
Determine the second quantity of the target range;
If second quantity is more than or equal to the second preset quantity threshold value, determine that the posture of the target user is different
Normal posture;
If second quantity is less than the second preset quantity threshold value, determine that the posture of the target user is non-exception
Posture.
Optionally, the described 4th determine that submodule is used for:
Obtain the corresponding distance weighting of key point described in every two;
Determine target range from key point distance, the target range be meet the key point of preset condition away from
From;
Posture is calculated according to the target range and the distance weighting and determines parameter;
If the posture determines that parameter is more than or equal to parameter preset threshold value, determine that the posture of the target user is different
Normal posture;
If the posture determines that parameter is less than parameter preset threshold value, determine that the posture of the target user is non-abnormal appearance
State.
Optionally, the key point includes multiple key points, and the determining module is used for:
Determine that the second target critical point, the second target critical point are any in the key point from the key point
Key point;
Reference frame is established centered on the second target critical point;
Other key points in the key point in addition to the second target critical point are mapped to the reference frame
In, obtain the relative position coordinates of other key points;
Using the relative position coordinates as the input of preset convolutional neural networks, to obtain the convolutional neural networks
The first probability and the second probability of output, first probability is that the posture of the target user is abnormal appearance probability of state, institute
Stating the posture that the second probability is the target user is non-abnormal appearance probability of state;
If first probability is greater than or equal to second probability, determine the posture of the target user for abnormal appearance
State;
If first probability is less than second probability, determine that the posture of the target user is non-abnormal posture.
Optionally, described device further include:
Alarm module, for if it is determined that the posture of the target user is abnormal posture, the first warning information of sending;With/
Or,
The alarm module, for if it is determined that the posture of the target user is abnormal posture, to the transmission of preset terminal
Second warning information.
Through the above technical solutions, target image of the disclosure by acquisition target user first, obtains target later and uses
The position coordinates of the key point at family in the target image, key point include the human body of pre-set target user, according to
Position coordinates determine whether the posture of target user is abnormal posture.The disclosure can not have apparent change in displacement in user
When, whether the posture for accurately determining user is abnormal posture, and detection efficiency is high, and does not need user and additionally carry detection dress
It sets, is easy to use, improves user experience.
Other feature and advantage of the disclosure will the following detailed description will be given in the detailed implementation section.
Detailed description of the invention
Attached drawing is and to constitute part of specification for providing further understanding of the disclosure, with following tool
Body embodiment is used to explain the disclosure together, but does not constitute the limitation to the disclosure.In the accompanying drawings:
Fig. 1 is a kind of flow chart of the detection method of abnormal posture shown according to an exemplary embodiment;
Fig. 2 is a kind of flow chart of step 103 shown in embodiment illustrated in fig. 1;
Fig. 3 is the signal that a kind of image collecting device shown in embodiment illustrated in fig. 1 is mounted on the wall of house side
Figure;
Fig. 4 is the flow chart of another step 103 shown in embodiment illustrated in fig. 1;
Fig. 5 is the schematic diagram that a kind of image collecting device shown in embodiment illustrated in fig. 1 is mounted at the top of house;
Fig. 6 is that another image collecting device shown in embodiment illustrated in fig. 1 is mounted on showing on the wall of house side
It is intended to;
Fig. 7 is the flow chart of another step 103 shown in embodiment illustrated in fig. 1;
Fig. 8 is the flow chart of the detection method of another abnormal posture shown according to an exemplary embodiment;
Fig. 9 is a kind of block diagram of the detection device of abnormal posture shown according to an exemplary embodiment;
Figure 10 is a kind of block diagram of determining module shown in embodiment illustrated in fig. 9;
Figure 11 is the block diagram of another determining module shown in embodiment illustrated in fig. 9;
Figure 12 is the block diagram of the detection device of another abnormal posture shown according to an exemplary embodiment.
Specific embodiment
Example embodiments are described in detail here, and the example is illustrated in the accompanying drawings.Following description is related to
When attached drawing, unless otherwise indicated, the same numbers in different drawings indicate the same or similar elements.Following exemplary embodiment
Described in embodiment do not represent all implementations consistent with this disclosure.On the contrary, they be only with it is such as appended
The example of the consistent device and method of some aspects be described in detail in claims, the disclosure.
Before introducing the detection method of abnormal posture of disclosure offer and device, first to each embodiment of the disclosure
Involved application scenarios are introduced.The detection method of the exception posture, which can be applied, old man, infant, physical disabilities etc.
Target user's movable scene is equipped with the image collecting device of acquisition two dimensional image in the scene, and image collecting device can
To be the devices such as camera, camera or infrared image sensor.Image collecting device, which for example may be mounted at the top of house, (to be made
Image collecting device can acquire downwards a certain range of target image), also may be mounted on the wall of house side (makes
Image collecting device can laterally acquire a certain range of target image).
Fig. 1 is a kind of flow chart of the detection method of abnormal posture shown according to an exemplary embodiment.Such as Fig. 1 institute
Show, method includes the following steps:
Step 101, the target image of target user is obtained.
Step 102, the position coordinates of the key point of target user in the target image are obtained, key point includes presetting
Target user human body.
Step 103, determine whether the posture of target user is abnormal posture according to position coordinates.
It is exemplary, abnormal attitude detection is carried out to target user, it is necessary first to obtain the target image of target user, target
What image can be image collecting device timing acquiring includes the image of target user.Target image is handled again, with
Obtain the corresponding position coordinates in the target image of key point of target user.If used in target image comprising multiple targets
Family can then obtain the position coordinates of the key point of each target user in multiple target users in the target image respectively.It is right
Target image handled there are many implementation, a kind of achievable mode be by depth convolutional neural networks (English:
DeepConvolutional Neural Network) target image is handled, it is used with the target in recognition target image
Family, and determine the corresponding key point of target user, so that it is determined that the position coordinates of key point in the target image.Depth convolution mind
It for example can be CPM (English: Convolutional PoseMachines) model through network, OpenPose model,
DeeperCut model etc..After getting position coordinates, determine whether the posture of target user is abnormal appearance according to position coordinates
State.Wherein, key point is one or more, including pre-set for determining whether the posture of target user is abnormal posture
Target user human body, can be the whole human bodies that can be identified in target image, be also possible to target image
One or more of specific human body.The key point of target user for example can be eyes, ear, nose, the corners of the mouth, neck
The human bodies such as son, shoulder, chest center, belly center, across portion center, knee joint, foot, elbow, hand.Abnormal posture for example may be used
To be posture that target user falls down or lies on the floor, non-exception posture for example can be the appearance that target user stands or sits
State.
It should be noted that can first obtain target user to improve the accuracy of abnormal attitude detection and stability
Multiple target images, later to each target image execute step 102 to step 103, determine in each target image
Whether the posture of target user is abnormal posture, and the posture for recording determining target user is the of the target image of abnormal posture
A kind of quantity, and determine that the posture of target user is the he second-class number amount of the target image of non-abnormal posture.Finally according to first
Whether class quantity and he second-class number amount are abnormal posture come the posture for determining target user.If such as first kind quantity is greater than or waits
In he second-class number amount, it is determined that the posture of target user is abnormal posture, otherwise determines that the posture of target user is non-abnormal appearance
State.
In conclusion the disclosure obtains the key of target user by the target image of acquisition target user first later
The position coordinates of point in the target image, key point includes the human body of pre-set target user, according to position coordinates
Whether the posture for determining target user is abnormal posture.The disclosure can be when user have apparent change in displacement, accurately
Whether the posture for determining user is abnormal posture, and detection efficiency is high, and does not need user and additionally carry detection device, convenient for making
With improving user experience.
Fig. 2 is a kind of flow chart of step 103 shown in embodiment illustrated in fig. 1.As shown in Fig. 2, step 103 includes following
Step:
Step 1031, the first image-region is determined from target image according to preset dividing line.
Step 1032, according to position coordinates and the first image-region, determine whether the posture of target user is abnormal posture.
Exemplary, the posture of target user is abnormal posture (such as target user falls down or lies on the floor) and target
The posture of user is non-abnormal posture (such as target user stands or sits), and certain key points of target user are in the target image
Locating region is different.For example, when image collecting device is mounted on the wall of house side, when the posture of target user is
When non-exception posture, locating region is higher in the target image for all key points of the upper part of the body of target user, and works as target
When the posture of user is abnormal posture, all key points of the upper part of the body of target user in the target image locating region compared with
It is low.Can preset in the target image a dividing line (such as can at the 0.4m of ground set one division
Line), and using the region between dividing line and ground as the first image-region, alternatively, dividing line in target image is below
Region is as the first image-region.A plurality of dividing line can also be preset in the target image, and a plurality of dividing line is surrounded
Region as the first image-region.
Using by the region below of dividing line in target image as the first image-region, as shown in figure 3, when target is used
When the posture at family is non-abnormal posture, all key points of the upper part of the body of target user all in dividing line more than, i.e., target is used
All key points of the family upper part of the body are not located in the first image-region, and when the posture of target user is become from non-abnormal posture
When abnormal posture, certain key points of the upper part of the body of target user can be in dividing line hereinafter, the upper part of the body of i.e. target user
Certain key points can be in the first image-region.It therefore, can be according to the position coordinates of key point in the target image and
The relationship of one image-region, to determine whether the posture of target user is abnormal posture.
Optionally, step 1032 can be accomplished by the following way:
A first object key point) is determined from key point, first object key point includes that position coordinates are located at the first image
Key point in region.
B the first quantity of first object key point) is determined.
C) if the first quantity is more than or equal to the first preset quantity threshold value, determine the posture of target user for abnormal appearance
State.
D) if the first quantity is less than the first preset quantity threshold value, determine that the posture of target user is non-abnormal posture.
Specifically, determining position first from key point when on the wall that image collecting device is mounted on house side
Coordinate is located at the first object key point (such as the certain key points that can be target user's upper part of the body) in the first image-region,
And determine the first quantity of first object key point.Later according to the size relation of the first quantity and the first preset quantity threshold value,
Whether the posture for determining target user is abnormal posture, if the first quantity is more than or equal to the first preset quantity threshold value, is determined
The posture of target user is abnormal posture, if the first quantity less than the first preset quantity threshold value, determines that the posture of target user is
Non- exception posture.For example, the first preset quantity threshold value is 2, the position of the key point of the target user got in the target image
Setting coordinate has 4, if the key point that position coordinates are located in the first image-region has 3, i.e. the of first object key point
One quantity is 3, is greater than the first preset quantity threshold value, it is determined that the posture of target user is abnormal posture, if position coordinates position
There is 1 in the key point in the first image-region, i.e. the first quantity of first object key point is 1, less than the first preset quantity
Threshold value, it is determined that the posture of target user is non-abnormal posture.
Fig. 4 is the flow chart of another step 103 shown in embodiment illustrated in fig. 1.As shown in figure 4, key point includes extremely
Few two key points, step 103 the following steps are included:
Step 1033, the key point distance between every two key point is determined according to position coordinates.
Step 1034, determine whether the posture of target user is abnormal posture according to key point distance.
Specifically, the posture of corresponding target user is different when the key point of target user includes at least two key point
Normal posture and the posture of target user are non-abnormal posture, and the position coordinates of key point in the target image are different.For example, when figure
When being mounted at the top of house as acquisition device, as shown in figure 5, a in Fig. 5 is that image collecting device is mounted on showing at the top of house
It is intended to, the b in Fig. 5 is the schematic diagram that image collecting device is mounted on target image when at the top of house, if the appearance of target user
State is non-abnormal posture, and the position coordinates of the key point of target user in the target image compare concentration (such as indicating ear
Piece, shoulder, hand, belly just neutralizing the position coordinates of kneed key point and comparing concentration), the key between every two key point
Point is closer, and if the posture of target user is abnormal posture, and the key point of target user is in the target image at this time
Position coordinates are more dispersed (such as indicating that ear, shoulder, hand, belly are just neutralizing the position coordinates of kneed key point
It is more dispersed), the key point distance between every two key point is farther out.When image collecting device is mounted on the wall of house side
When upper, as shown in fig. 6, a in Fig. 6 is that image collecting device is mounted on the schematic diagram on the wall of house side, the b in Fig. 6
The schematic diagram of target image when on the wall of house side is mounted on for image collecting device, if the posture of target user is non-
Abnormal posture, the position coordinates of the key point of target user in the target image it is more dispersed (such as indicate shoulder, hand,
Chest center, knee joint and foot key point position coordinates it is more dispersed), key point distance between every two key point compared with
Far, but when the posture of target user is abnormal posture, the position coordinates of the key point of target user in the target image at this time
Compare concentration (such as indicate to indicate shoulder, hand, chest center, knee joint and foot the position coordinates of key point compare collection
In), the key point between every two key point is closer.Therefore, every two key point first can be determined according to position coordinates
Between key point distance, further according to key point distance determine target user posture whether be abnormal posture.
Optionally, step 1034 can be realized by following steps:
1) target range is determined from key point distance, target range is to meet the key point distance of preset condition.
For example, determine whether the posture of target user is abnormal posture according to key point distance, it first can be from pass
Target range is determined in key point distance, wherein target range is to meet the key point distance of preset condition.Work as image collecting device
When being mounted at the top of house, preset condition for example can be key point distance and be greater than or equal to the corresponding distance of every two key point
Threshold value, when image collecting device is mounted on the wall of house side, preset condition for example can be key point distance and be less than
Or it is equal to the corresponding distance threshold of every two key point.For on the wall that image collecting device is mounted on house side, because
For range usually less (such as the distance between the target user and image collecting device of image acquisition device target image
Less than 5m), it, should be significantly big in the key point distance of the target user of the non-abnormal posture of range image acquisition device farthest
In the key point distance of the target user of the abnormal posture of range image acquisition device most nearby.It therefore, can be by range image
The key point distance of the target user of the non-abnormal posture of acquisition device farthest is corresponding apart from threshold as every two key point
Value.
2) the second quantity of target range is determined.
If 3) the second quantity is more than or equal to the second preset quantity threshold value, determine the posture of target user for abnormal appearance
State.
4) if the second quantity is less than the second preset quantity threshold value, determine that the posture of target user is non-abnormal posture.
Further, after determining target range in key point distance, the second quantity of target range is determined, further according to the
The relationship of two quantity and the second preset quantity threshold value determines whether the posture of target user is abnormal posture.If the second quantity is big
In or equal to the second preset quantity threshold value, the posture of target user is determined for abnormal posture, if the second quantity is pre- less than second
If amount threshold, determine that the posture of target user is non-abnormal posture.For example, the second preset quantity threshold value is 3, target user's
The position coordinates of key point in the target image have 4, then the key between every two key point is determined according to 4 position coordinates
Point distance has 6, if the key point distance for meeting preset condition has 4, i.e., the second quantity of target range is 4, is greater than the
Two preset quantity threshold values, it is determined that the posture of target user is abnormal posture, has 2 if meeting the key point distance of preset condition
A, i.e., the second quantity of target range is 2, less than the second preset quantity threshold value, it is determined that the posture of target user is non-exception
Posture.
In another implementation, step 1032 can be realized by following steps:
A) the corresponding distance weighting of every two key point is obtained.
B) target range is determined from key point distance, target range is to meet the key point distance of preset condition.
C) posture is calculated according to target range and distance weighting and determines parameter.
If d) posture determines that parameter is more than or equal to parameter preset threshold value, determine the posture of target user for abnormal appearance
State.
If e) posture determines that parameter is less than parameter preset threshold value, determine that the posture of target user is non-abnormal posture.
Another scene is become from non-abnormal posture the mistake of abnormal posture in the posture of target user in a practical situation
Cheng Zhong, certain key point distance changes it is larger (such as the key point for indicating nose and the key point for indicating crotch it
Between key point distance change it is larger), and some key point distance changes are (such as two for indicating eyes smaller or constant
Key point distance change between key point is smaller or constant), key point distance change is bigger, determines that the posture of target user is
No is that abnormal posture is also more accurate.That is, different key point distances, determines whether the posture of target user is abnormal
The significance level of posture is also different.Therefore, can according to different key point distances to determine target user posture whether be
The significance level of abnormal posture, to preset different key points apart from corresponding distance weighting.According to key point distance
During determining whether the posture of target user is abnormal posture, the corresponding distance power of every two key point can be first obtained
Weight determines the target range for meeting preset condition from key point distance.It is calculated later according to target range and distance weighting
Parameter is determined to posture, if posture determines that parameter is more than or equal to parameter preset threshold value, determines that the posture of target user is different
Normal posture determines that the posture of target user is non-abnormal posture if posture determines that parameter is less than parameter preset threshold value.When image is adopted
When acquisition means are mounted at the top of house, it is corresponding more than or equal to every two key point that preset condition for example can be key point distance
Distance threshold, when image collecting device is mounted on the wall of house side, preset condition for example can be key point away from
From less than or equal to the corresponding distance threshold of every two key point.
Fig. 7 is the flow chart of another step 103 shown in embodiment illustrated in fig. 1.As shown in fig. 7, key point includes more
A key point, step 103 the following steps are included:
Step 1035, determine that the second target critical point, the second target critical point are any pass in key point from key point
Key point.
Step 1036, reference frame is established centered on the second target critical point.
Step 1037, other key points in key point in addition to the second target critical point are mapped in reference frame,
Obtain the relative position coordinates of other key points.
Step 1038, using relative position coordinates as the input of preset convolutional neural networks, to obtain convolutional Neural net
The first probability and the second probability of network output, the first probability is that the posture of target user is abnormal appearance probability of state, the second probability
Posture for target user is non-abnormal appearance probability of state.
Step 1039, if the first probability is greater than or equal to the second probability, determine the posture of target user for abnormal posture.
Step 1030, if the first probability is less than the second probability, determine that the posture of target user is non-abnormal posture.
It, can also be using position coordinates as the input of preset convolutional neural networks, to determine mesh in another scene
Whether the posture for marking user is abnormal posture.Determine that the second target critical point, the second target critical point are first from key point
Any key point in key point, and reference frame is established centered on the second target critical point, then second will be removed in key point
Other key points outside target critical point are mapped in reference frame, obtain the relative position coordinates of other key points.Later
Using the relative position coordinates of other key points as the input of convolutional neural networks, to obtain the first of convolutional neural networks output
Probability and the second probability, wherein the first probability is that the posture of target user is abnormal appearance probability of state, and the second probability is target use
The posture at family is non-abnormal appearance probability of state.Preset convolutional neural networks for example can be MLP (English: Multilayer
Perceptron, Chinese: multi-layer perception (MLP)) neural network.Finally determined according to the size relation of the first probability and the second probability
Whether the posture of target user is non-abnormal posture, if the first probability determines the appearance of target user more than or equal to the second probability
State is abnormal posture, if the first probability less than the second probability, determines that the posture of target user is non-abnormal posture.
Fig. 8 is the flow chart of the detection method of another abnormal posture shown according to an exemplary embodiment.Such as Fig. 8 institute
Show, this method is further comprising the steps of:
Step 104, however, it is determined that the posture of target user is abnormal posture, issues the first warning information.And/or
Step 105, however, it is determined that the posture of target user is abnormal posture, sends the second warning information to preset terminal.
It is exemplary, however, it is determined that the posture of target user is abnormal posture, illustrates that target user is likely to be at dangerous shape at this time
State can issue the first warning information, and/or, the second warning information is sent to preset terminal, so that target user endangers
It can be succoured in time when dangerous.The mode for issuing the first warning information can for example show that the first alarm is believed on a display screen
Breath, be also possible to be flashed by control instructions lamp according to preset mode (such as control instructions lamp is according to preset frequency
Flashed with color), it can also be and voice prompting is issued by control loudspeaker.Preset terminal can be pre-set
Terminal (such as: the terminal that the Related Contact of target user uses) or target user region with target user's binding
Hospital provisioned in terminal, wherein terminal can be smart phone, tablet computer, smartwatch, Intelligent bracelet, PDA (English
Text: Personal Digital Assistant, Chinese: personal digital assistant) etc. mobile terminals, be also possible to desktop computer
Equal fixed terminals.
In conclusion the disclosure obtains the key of target user by the target image of acquisition target user first later
The position coordinates of point in the target image, key point includes the human body of pre-set target user, according to position coordinates
Whether the posture for determining target user is abnormal posture.The disclosure can be when user have apparent change in displacement, accurately
Whether the posture for determining user is abnormal posture, and detection efficiency is high, and does not need user and additionally carry detection device, convenient for making
With improving user experience.
Fig. 9 is a kind of block diagram of the detection device of abnormal posture shown according to an exemplary embodiment.As shown in figure 9,
The device 200 includes:
Module 201 is obtained, for obtaining the target image of target user.
Module 201 is obtained, is also used to obtain the position coordinates of the key point of target user in the target image, key point packet
Include the human body of pre-set target user.
Determining module 202, for determining whether the posture of target user is abnormal posture according to position coordinates.
Figure 10 is a kind of block diagram of determining module shown in embodiment illustrated in fig. 9.As shown in Figure 10, determining module 202 is wrapped
It includes:
First determines submodule 2021, for determining the first image-region from target image according to preset dividing line.
Second determines submodule 2022, for determining the posture of target user according to position coordinates and the first image-region
It whether is abnormal posture.
Optionally, second determine that submodule 2022 is used for:
First object key point is determined from key point, first object key point includes that position coordinates are located at the first image district
Key point in domain.
Determine the first quantity of first object key point.
If the first quantity is more than or equal to the first preset quantity threshold value, determine the posture of target user for abnormal posture.
If the first quantity less than the first preset quantity threshold value, determines that the posture of target user is non-abnormal posture.
Figure 11 is a kind of block diagram of determining module shown in embodiment illustrated in fig. 9.As shown in figure 11, key point includes at least
Two key points, determining module 202 include:
Third determines submodule 2023, for determining the key point distance between every two key point according to position coordinates.
4th determines submodule 2024, for determining whether the posture of target user is abnormal appearance according to key point distance
State.
Optionally, the 4th determine that submodule 2024 is used for:
Target range is determined from key point distance, and target range is to meet the key point distance of preset condition.
Determine the second quantity of target range.
If the second quantity is more than or equal to the second preset quantity threshold value, determine the posture of target user for abnormal posture.
If the second quantity less than the second preset quantity threshold value, determines that the posture of target user is non-abnormal posture.
Optionally, the 4th determine that submodule 2024 is used for:
Obtain the corresponding distance weighting of every two key point.
Target range is determined from key point distance, and target range is to meet the key point distance of preset condition.
Posture is calculated according to target range and distance weighting and determines parameter.
If posture determines that parameter is more than or equal to parameter preset threshold value, determine the posture of target user for abnormal posture.
If posture determines that parameter is less than parameter preset threshold value, determine that the posture of target user is non-abnormal posture.
Optionally, key point includes multiple key points, and determining module 202 is used for:
Determine that the second target critical point, the second target critical point are any key point in key point from key point.
Reference frame is established centered on the second target critical point.
Other key points in key point in addition to the second target critical point are mapped in reference frame, other passes are obtained
The relative position coordinates of key point.
Using relative position coordinates as the input of preset convolutional neural networks, to obtain the of convolutional neural networks output
One probability and the second probability, the first probability are that the posture of target user is abnormal appearance probability of state, and the second probability is target user
Posture be non-abnormal appearance probability of state.
If the first probability is greater than or equal to the second probability, determine the posture of target user for abnormal posture.
If the first probability less than the second probability, determines that the posture of target user is non-abnormal posture.
Figure 12 is the block diagram of the detection device of another abnormal posture shown according to an exemplary embodiment.Such as Figure 12 institute
Show, device 200 further include:
Alarm module 203, for if it is determined that the posture of target user is abnormal posture, the first warning information of sending.With/
Or,
Alarm module 203, for if it is determined that the posture of target user is abnormal posture, to the second announcement of preset terminal transmission
Alert information.
About the device in above-described embodiment, wherein modules execute the concrete mode of operation in related this method
Embodiment in be described in detail, no detailed explanation will be given here.
In conclusion the disclosure obtains the key of target user by the target image of acquisition target user first later
The position coordinates of point in the target image, key point includes the human body of pre-set target user, according to position coordinates
Whether the posture for determining target user is abnormal posture.The disclosure can be when user have apparent change in displacement, accurately
Whether the posture for determining user is abnormal posture, and detection efficiency is high, and does not need user and additionally carry detection device, convenient for making
With improving user experience.
The preferred embodiment of the disclosure is described in detail in conjunction with attached drawing above, still, the disclosure is not limited to above-mentioned reality
The detail in mode is applied, in the range of the technology design of the disclosure, a variety of letters can be carried out to the technical solution of the disclosure
Monotropic type, these simple variants belong to the protection scope of the disclosure.
It is further to note that specific technical features described in the above specific embodiments, in not lance
In the case where shield, can be combined in any appropriate way, in order to avoid unnecessary repetition, the disclosure to it is various can
No further explanation will be given for the combination of energy.
In addition, any combination can also be carried out between a variety of different embodiments of the disclosure, as long as it is without prejudice to originally
Disclosed thought equally should be considered as disclosure disclosure of that.
Claims (10)
1. a kind of detection method of exception posture, which is characterized in that the described method includes:
Obtain the target image of target user;
Position coordinates of the key point of the target user in the target image are obtained, the key point includes presetting
The target user human body;
Whether the posture that the target user is determined according to the position coordinates is abnormal posture.
2. the method according to claim 1, wherein described determine the target user according to the position coordinates
Posture whether be that abnormal posture includes:
The first image-region is determined from the target image according to preset dividing line;
According to the position coordinates and the first image region, determine whether the posture of the target user is abnormal posture.
3. according to the method described in claim 2, it is characterized in that, described according to the position coordinates and the first image area
Domain determines whether the posture of the target user is that abnormal posture includes:
First object key point is determined from the key point, the first object key point includes that the position coordinates are located at institute
State the key point in the first image-region;
Determine the first quantity of the first object key point;
If first quantity is more than or equal to the first preset quantity threshold value, determine the posture of the target user for abnormal appearance
State;
If first quantity is less than the first preset quantity threshold value, determine that the posture of the target user is non-abnormal appearance
State.
4. the method according to claim 1, wherein the key point includes at least two key points, institute
It states and determines whether the posture of the target user is that abnormal posture includes: according to the position coordinates
The key point distance between key point described in every two is determined according to the position coordinates;
Determine whether the posture of the target user is abnormal posture according to the key point distance.
5. according to the method described in claim 4, it is characterized in that, described determine that the target is used according to the key point distance
Whether the posture at family is that abnormal posture includes:
Target range is determined from the key point distance, and the target range is to meet the key point distance of preset condition;
Determine the second quantity of the target range;
If second quantity is more than or equal to the second preset quantity threshold value, determine the posture of the target user for abnormal appearance
State;
If second quantity is less than the second preset quantity threshold value, determine that the posture of the target user is non-abnormal appearance
State.
6. according to the method described in claim 4, it is characterized in that, described determine that the target is used according to the key point distance
Whether the posture at family is that abnormal posture includes:
Obtain the corresponding distance weighting of key point described in every two;
Target range is determined from the key point distance, and the target range is to meet the key point distance of preset condition;
Posture is calculated according to the target range and the distance weighting and determines parameter;
If the posture determines that parameter is more than or equal to parameter preset threshold value, determine the posture of the target user for abnormal appearance
State;
If the posture determines that parameter is less than parameter preset threshold value, determine that the posture of the target user is non-abnormal posture.
7. the method according to claim 1, wherein the key point includes multiple key points, described
Whether the posture that the target user is determined according to the position coordinates is that abnormal posture includes:
Determine that the second target critical point, the second target critical point are any key in the key point from the key point
Point;
Reference frame is established centered on the second target critical point;
Other key points in the key point in addition to the second target critical point are mapped in the reference frame, are obtained
To the relative position coordinates of other key points;
Using the relative position coordinates as the input of preset convolutional neural networks, to obtain the convolutional neural networks output
The first probability and the second probability, first probability is that the posture of the target user is abnormal appearance probability of state, described
Two probability are that the posture of the target user is non-abnormal appearance probability of state;
If first probability is greater than or equal to second probability, determine the posture of the target user for abnormal posture;
If first probability is less than second probability, determine that the posture of the target user is non-abnormal posture.
8. method according to any one of claim 1 to 7, which is characterized in that the method also includes:
If it is determined that the posture of the target user is abnormal posture, the first warning information is issued;And/or
If it is determined that the posture of the target user is abnormal posture, the second warning information is sent to preset terminal.
9. a kind of detection device of exception posture, which is characterized in that described device includes:
Module is obtained, for obtaining the target image of target user;
The acquisition module is also used to obtain position coordinates of the key point of the target user in the target image, institute
State the human body that key point includes the pre-set target user;
Determining module, for determining whether the posture of the target user is abnormal posture according to the position coordinates.
10. device according to claim 9, which is characterized in that the determining module includes:
First determines submodule, for determining the first image-region from the target image according to preset dividing line;
Second determines submodule, for determining the target user's according to the position coordinates and the first image region
Whether posture is abnormal posture.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910471561 | 2019-05-31 | ||
CN2019104715612 | 2019-05-31 |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110378244A true CN110378244A (en) | 2019-10-25 |
CN110378244B CN110378244B (en) | 2021-12-31 |
Family
ID=68249418
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910562307.3A Active CN110378244B (en) | 2019-05-31 | 2019-06-26 | Abnormal posture detection method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110378244B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111125768A (en) * | 2019-12-26 | 2020-05-08 | 联想(北京)有限公司 | Information processing method, information processing apparatus, electronic device, and medium |
CN111460886A (en) * | 2020-02-21 | 2020-07-28 | 珠海格力电器股份有限公司 | Monitoring method and device and computer equipment |
CN111767888A (en) * | 2020-07-08 | 2020-10-13 | 北京澎思科技有限公司 | Object state detection method, computer device, storage medium, and electronic device |
CN112613440A (en) * | 2020-12-29 | 2021-04-06 | 北京市商汤科技开发有限公司 | Attitude detection method and apparatus, electronic device and storage medium |
CN112674759A (en) * | 2020-12-21 | 2021-04-20 | 西南交通大学 | Baby standing state identification method and system |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108764185A (en) * | 2018-06-01 | 2018-11-06 | 京东方科技集团股份有限公司 | A kind of image processing method and device |
CN109241218A (en) * | 2018-08-17 | 2019-01-18 | 北京云鸟科技有限公司 | A kind of data point indication method and device |
CN109389640A (en) * | 2018-09-29 | 2019-02-26 | 北京字节跳动网络技术有限公司 | Image processing method and device |
CN109508576A (en) * | 2017-09-14 | 2019-03-22 | 杭州海康威视数字技术股份有限公司 | A kind of abnormal driving behavioral value method, apparatus and electronic equipment |
CN109919132A (en) * | 2019-03-22 | 2019-06-21 | 广东省智能制造研究所 | A kind of pedestrian's tumble recognition methods based on skeleton detection |
-
2019
- 2019-06-26 CN CN201910562307.3A patent/CN110378244B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109508576A (en) * | 2017-09-14 | 2019-03-22 | 杭州海康威视数字技术股份有限公司 | A kind of abnormal driving behavioral value method, apparatus and electronic equipment |
CN108764185A (en) * | 2018-06-01 | 2018-11-06 | 京东方科技集团股份有限公司 | A kind of image processing method and device |
CN109241218A (en) * | 2018-08-17 | 2019-01-18 | 北京云鸟科技有限公司 | A kind of data point indication method and device |
CN109389640A (en) * | 2018-09-29 | 2019-02-26 | 北京字节跳动网络技术有限公司 | Image processing method and device |
CN109919132A (en) * | 2019-03-22 | 2019-06-21 | 广东省智能制造研究所 | A kind of pedestrian's tumble recognition methods based on skeleton detection |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111125768A (en) * | 2019-12-26 | 2020-05-08 | 联想(北京)有限公司 | Information processing method, information processing apparatus, electronic device, and medium |
CN111460886A (en) * | 2020-02-21 | 2020-07-28 | 珠海格力电器股份有限公司 | Monitoring method and device and computer equipment |
CN111460886B (en) * | 2020-02-21 | 2024-01-16 | 珠海格力电器股份有限公司 | Monitoring method, device and computer equipment |
CN111767888A (en) * | 2020-07-08 | 2020-10-13 | 北京澎思科技有限公司 | Object state detection method, computer device, storage medium, and electronic device |
CN112674759A (en) * | 2020-12-21 | 2021-04-20 | 西南交通大学 | Baby standing state identification method and system |
CN112613440A (en) * | 2020-12-29 | 2021-04-06 | 北京市商汤科技开发有限公司 | Attitude detection method and apparatus, electronic device and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN110378244B (en) | 2021-12-31 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110378244A (en) | The detection method and device of abnormal posture | |
CN103955272B (en) | A kind of terminal user's attitude detection system | |
CN109961058B (en) | Non-contact fall detection method and device | |
RU2584459C2 (en) | Gesture control for monitoring vital body signs | |
CN110472481B (en) | Sleeping gesture detection method, device and equipment | |
KR102548453B1 (en) | Electronic device and controling method thereof | |
CN106095421B (en) | A kind of control method and system of virtual reality device | |
JP2021064367A (en) | Motion recognition method, motion recognition device, and electronic device | |
TWI680440B (en) | Image detection method and image detection device for determining postures of user | |
EP4062824A1 (en) | Electrocardiosignal prediction method and device, terminals, and storage medium | |
CN109711289B (en) | Riding reminding method and device, electronic equipment and storage medium | |
CN113384267A (en) | Fall real-time detection method, system, terminal equipment and storage medium | |
CN117503099A (en) | Human body composition analyzer, measuring method, measuring device, apparatus and storage medium | |
CN205107652U (en) | Intelligent wearable device | |
KR101641643B1 (en) | Method and Apparatus for Body State Recognition using the Impedance Charateristic Enhanced Electorde | |
CN108846996B (en) | Tumble detection system and method | |
CN203930836U (en) | Wearable electronic equipment | |
JP2010176588A (en) | Terminal device, image processing method, and program | |
CN109345636A (en) | The method and apparatus for obtaining conjecture face figure | |
CN113743234A (en) | Target action determining method, target action counting method and electronic device | |
CN109547916B (en) | Intelligent lighting equipment, communication system and indoor positioning method | |
CN111150318B (en) | Liquid soap discharging control method and related device | |
JP7449212B2 (en) | Systems, electronic devices, control methods for electronic devices, and programs | |
CN107346600A (en) | Positioning falling alarm takes | |
TW202011262A (en) | Image detection method and image detection device utilizing dual analysis |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |