One kind falls down alarm method, device and equipment
Technical field
The present invention relates to falling-resistant monitoring field, alarm method, device and equipment are fallen down in particular to one kind.
Background technology
According to world's disease control and prevention tissue statistics, the old man more than 65 years old, has l/3 to fall down every year in the world,
Half of which is fallen down for recidivity, is fallen down rate and is increased with the age and increase;There is 20%-30% human hair in the elderly fallen down
Raw moderate and severe injury (hip fracture and head wound), 60% limitation of activity or stiff causes huge medical burden
And health hazards.In hospital, patient, which falls down to give treatment to not in time, can also produce very serious consequence.Reduce the elderly and patient falls
The injury problem that rewinding comes is as study hotspot new in the world.Medical research is indicated:Reduce when people falls down and brought to body
Impact can effectively mitigate the injury that the accident of falling down is brought;And old man and patient for having fallen down, succour the stand-by period
Length directly determine illness degree.
Automatic detection alarming system for falling over main at present mainly passes through the acceleration of metastomium on automatic detection human body
Information and posture information, the relation between comprehensive human body acceleration, pose, run duration three, correctly judge whether human body falls
And whether need emergency.
However, it is necessary to wear extra equipment this restriction, the effect of realizing of existing scheme, and equipment are hindered significantly
Wear relatively complicated, the elderly or patient were not in contact with not easy to operate if such device before this, and equipment wearing position has
It is strict with, ordinary people is difficult to grasp, and additionally professional may be needed to on-site install.In addition, the elderly or patient take action mostly
Inconvenience, it is likely that occur because health is not in time for the situation of wearable device, now once dangerous occur, it is impossible in time
Alarm, consequence will be hardly imaginable, if to ensure system worked well in this case, and human body needs to wear for 24 hours to set
It is standby, it have impact on the level of comfort of the elderly and patient.
The content of the invention
Alarm method, device and equipment are fallen down it is an object of the invention to provide one kind, to improve above mentioned problem.
To achieve these goals, the technical solution adopted by the present invention is as follows:
In a first aspect, one kind falls down alarm method, the alarm method of falling down includes:
Obtain the ambient video in pickup area;
The ambient video is processed into the image sequence of multiframe first;
Each frame described first image sequence inputting is estimated into model to the human body attitude based on deep learning algorithm, obtained
The corresponding temperature image of human body in the ambient video;
Multiple key points of the temperature image are obtained, the multiple key point is generated into the first key point by preset rules
Sequence;
Judge that the human body is according to the corresponding described first crucial point sequence of the continuous described first image sequence of multiframe
It is no to fall down;
When judging the falling over of human body, warning message is generated.
Further, it is described by each frame described first image sequence inputting to the human body attitude based on deep learning algorithm
Estimate model, before the step of obtaining the corresponding temperature image of human body in the ambient video, in addition to depth is based on to described
The step of human body attitude estimation model of learning algorithm is trained:
Human body picture is obtained, the profile of the human body in picture is labeled;
Build deep learning network;
The human body picture is inputted into the deep learning network to be trained, it is described based on deep learning algorithm to obtain
Human body attitude estimation model.
Further, it is described to be sentenced according to the corresponding described first crucial point sequence of the continuous described first image sequence of multiframe
The step of also including setting up key point series model before the step of whether human body that breaks is fallen down, the crucial point sequence of the foundation
The step of model, includes:
Human body corresponding second image sequence in different postures is obtained, second image sequence corresponding second is obtained
Crucial point sequence, the posture includes falling down behavior and does not fall down behavior;
The described second crucial point sequence is marked according to the corresponding posture of the second image sequence, described second is obtained and closes
The corresponding human body attitude change of key point sequence, to obtain the key point series model.
Further, acquisition human body corresponding second image sequence in different postures, obtains second image
The step of sequence corresponding second crucial point sequence, including:
Human body attitude model video is obtained, by the Video processing into the second image sequence described in multiframe;
Obtain the described second crucial point sequence of continuous second image sequence of multiframe.
Further, the step of the second crucial point sequence of acquisition multiframe continuous second image sequence includes:
Record the corresponding time of each second image sequence in continuous second image sequence of multiframe;
Each second image sequence corresponding second in continuous second image sequence of the multiframe is obtained to close
Key point sequence.
Further, it is described to be sentenced according to the corresponding described first crucial point sequence of the continuous described first image sequence of multiframe
The step of whether human body that breaks is fallen down, including:
When the change of the corresponding described first crucial point sequence of continuous multiframe described first image sequence is more than continuously
Judge the falling over of human body during threshold value of the change of the corresponding described second crucial point sequence of the second image sequence described in multiframe.
Second aspect, one kind falls down warning device, and the warning device of falling down includes:
Video acquiring module, for obtaining the ambient video in pickup area;
Image segmentation module, for the ambient video to be processed into the image sequence of multiframe first;
Temperature image collection module, for by each frame described first image sequence inputting to based on deep learning algorithm
Human body attitude estimates model, obtains the corresponding temperature image of human body in the ambient video;
Critical point detection module, multiple key points for obtaining the temperature image, by the multiple key point by pre-
If the crucial point sequence of rule generation first;
Behavior judge module is fallen down, for according to corresponding first key of the continuous described first image sequence of multiframe
Point sequence judges whether the human body is fallen down;
Alarm module, for when judging the falling over of human body, generating warning message.
Further, the warning device also includes:
Human body estimates model training module, and for obtaining human body picture, the profile of the human body in picture is labeled;Structure
Build deep learning network;The human body picture is inputted into the deep learning network to be trained, it is described based on depth to obtain
The human body attitude estimation model of learning algorithm.
Further, the warning device also includes:
Key point series model module, for obtaining human body corresponding second image sequence in different postures, obtains institute
The corresponding second crucial point sequence of the second image sequence is stated, the posture includes falling down behavior and do not fall down behavior;According to second
Described second crucial point sequence is marked the corresponding posture of image sequence, obtains the corresponding people of the second key point sequence
Body attitudes vibration, to obtain the key point series model.
The third aspect, one kind falls down warning device, and the warning device of falling down includes:
Memory;
Processor;And
Warning device is fallen down, the warning device of falling down is stored in the memory and including one or more by described
The software function module of computing device, falling down warning device includes:
Video acquiring module, for obtaining the ambient video in pickup area;
Image segmentation module, for the ambient video to be processed into the image sequence of multiframe first;
Temperature image collection module, for by each frame described first image sequence inputting to based on deep learning algorithm
Human body attitude estimates model, obtains the corresponding temperature image of human body in the ambient video;
Critical point detection module, multiple key points for obtaining the temperature image, by the multiple key point by pre-
If the crucial point sequence of rule generation first;
Behavior judge module is fallen down, for according to corresponding first key of the continuous described first image sequence of multiframe
Point sequence judges whether the human body is fallen down;
Alarm module, for when judging the falling over of human body, generating warning message;
Human body estimates model training module, for obtaining human body picture, the profile of the human body in picture is labeled, structure
Deep learning network is built, the human body picture is inputted into the deep learning network is trained, it is described based on depth to obtain
The human body attitude estimation model of learning algorithm;
Key point series model module, for obtaining human body corresponding second image sequence in different postures, obtains institute
The corresponding second crucial point sequence of the second image sequence is stated, the posture includes falling down behavior and do not fall down behavior;According to second
Described second crucial point sequence is marked the corresponding posture of image sequence, obtains the corresponding people of the second key point sequence
Body attitudes vibration, to obtain the key point series model.
Alarm method, device and equipment are fallen down the invention provides one kind, this method is applied to the device, and the equipment includes
The device, this method includes:Obtain the ambient video in pickup area;Ambient video is processed into the image sequence of multiframe first;
Each image sequence of frame first is input to the human body attitude based on deep learning algorithm and estimates model, people in ambient video is obtained
The corresponding temperature image of body;Multiple key points of temperature image are obtained, multiple key points are crucial by preset rules generation first
Point sequence;Judge whether human body is fallen down according to the corresponding first crucial point sequence of continuous first image sequence of multiframe;Work as judgement
During falling over of human body, warning message is generated.By gathering the human body in ambient video, the change of human body attitude is obtained, people is being judged
Warning message is generated when body is fallen down, is worn without human body, does not influence to detect body state while physical activity.
To enable the above objects, features and advantages of the present invention to become apparent, preferred embodiment cited below particularly, and coordinate
Appended accompanying drawing, is described in detail below.
Brief description of the drawings
In order to illustrate the technical solution of the embodiments of the present invention more clearly, below will be attached to what is used required in embodiment
Figure is briefly described, it will be appreciated that the following drawings illustrate only certain embodiments of the present invention, therefore is not construed as pair
The restriction of scope, for those of ordinary skill in the art, on the premise of not paying creative work, can also be according to this
A little accompanying drawings obtain other related accompanying drawings.
Fig. 1 is the schematic diagram provided in an embodiment of the present invention that interact for falling down warning device and multiple camera devices;
Fig. 2 is the block diagram provided in an embodiment of the present invention for falling down warning device;
Fig. 3-Fig. 5, is the high-level schematic functional block diagram provided in an embodiment of the present invention for falling down warning device;
Fig. 6 falls down the flow chart that alarm method is applied to fall down warning device to be provided in an embodiment of the present invention;
Fig. 7 sets up the step of human body attitude estimates model to be provided in an embodiment of the present invention;
Fig. 8-Figure 10 for it is provided in an embodiment of the present invention set up key point series model the step of.
Icon:100- falls down warning device;110- falls down warning device;112- video acquiring modules;113- images are split
Module;114- temperature image collection modules;115- critical point detection modules;116- falls down behavior judge module;117- alarm moulds
Block;118- human bodies estimate model training module;119- key point series model modules;120- memories;130- processors;140-
Communication unit;200- camera devices;300- networks.
Embodiment
To make the purpose, technical scheme and advantage of the embodiment of the present invention clearer, below in conjunction with the embodiment of the present invention
In accompanying drawing, the technical scheme in the embodiment of the present invention is clearly and completely described, it is clear that described embodiment is
A part of embodiment of the present invention, rather than whole embodiments.The present invention implementation being generally described and illustrated herein in the accompanying drawings
The component of example can be arranged and designed with a variety of configurations.
Therefore, the detailed description of embodiments of the invention below to providing in the accompanying drawings is not intended to limit claimed
The scope of the present invention, but be merely representative of the present invention selected embodiment.Based on the embodiment in the present invention, this area is common
The every other embodiment that technical staff is obtained under the premise of creative work is not made, belongs to the model that the present invention is protected
Enclose.
It should be noted that:Similar label and letter represents similar terms in following accompanying drawing, therefore, once a certain Xiang Yi
It is defined in individual accompanying drawing, then it further need not be defined and explained in subsequent accompanying drawing.
Referring to Fig. 1, falling down interacting for warning device 100 and multiple camera devices 200 to be provided in an embodiment of the present invention
Schematic diagram.Falling down warning device 100 can be communicated by network 300 with multiple camera devices 200, and alarm is fallen down to realize
Data communication between equipment 100 and multiple camera devices 200 is interacted.
Referring to Fig. 2, being the block diagram provided in an embodiment of the present invention for falling down warning device 100.Alarm is fallen down to set
Standby 100 include falling down warning device 110, memory 120, processor 130 and communication unit 140.Memory 120, processor 130
And each element of communication unit 140 is directly or indirectly electrically connected with each other, to realize the transmission or interaction of data.Example
Such as, these elements can be realized by one or more communication bus or signal wire be electrically connected with each other.Fall down warning device
110, which include at least one, can be stored in memory 120 or be solidificated in the form of software or firmware (firmware) and fall down report
Software function module in the operating system (operating system, OS) of alert equipment 100.Processor 130 is deposited for execution
The executable module stored in reservoir 120, for example, fall down the software function module and computer program included by warning device 110
Deng.
Wherein, the memory 120 may be, but not limited to, random access memory (Random Access
Memory, RAM), read-only storage (Read Only Memory, ROM), programmable read only memory (Programmable
Read-Only Memory, PROM), erasable read-only memory (Erasable Programmable Read-Only
Memory, EPROM), electricallyerasable ROM (EEROM) (Electric Erasable Programmable Read-Only
Memory, EEPROM) etc..Wherein, memory 120 is used for storage program, and processor 130 is performed after execute instruction is received
Described program.Communication unit 140, which is used to set up by network 300, to be fallen down between warning device 100 and multiple camera devices 200
Communication connection, and for passing through the transceiving data of network 300.
Referring to Fig. 3, being the high-level schematic functional block diagram provided in an embodiment of the present invention for falling down warning device 110.Fall down report
Alarm device 110 includes:Video acquiring module 112, image segmentation module 113, temperature image collection module 114, critical point detection
Module 115, fall down behavior judge module 116 and alarm module 117.
Wherein,
Video acquiring module 112, for obtaining the ambient video in pickup area.
In the present embodiment, video acquiring module 112 is used for the video information for obtaining the collection of camera device 200.The shooting is filled
It can be, installed in solitary old camera in other and in hospital ward and corridor, to gather real-time image information to put 200, on
Reach and fall down the progress subsequent treatment of warning device 100.Camera, which is installed, to be ensured, without monitoring dead angle, with this to ensure once falling down row
To occur energy and alarm, notify that associated care personnel are given treatment to., it will be clear that the ambient video can be shooting dress
When putting 200 and collecting human body and occur, human body is followed the trail of and shoots obtained video.
Image segmentation module 113, for ambient video to be processed into the image sequence of multiframe first.
In the present embodiment, the video that image segmentation module 113 is used to obtain video acquiring module 112 carries out segmentation portion
Reason, obtains multiple continuous image informations, and image information includes image., it will be clear that can be according to specifically using field
Scape, sets extraction multiple image per second.For example, being set to 6 two field pictures of interior extraction per second, it is of course possible to for specific required precision
Number of image frames is extracted in change.
Temperature image collection module 114, for each image sequence of frame first to be input to based on deep learning algorithm
Human body attitude estimates model, obtains the corresponding temperature image of human body in ambient video.
In the present embodiment, the image that temperature image collection module 114 is used to divide the image into the segmentation of module 113 is input to base
Estimate model in the human body attitude of deep learning algorithm, generate corresponding temperature figure.When it is implemented, based on deep learning algorithm
Human body attitude estimation model use deep learning framework Caffe (Convolutional Architecture for
Fast Feature Embedding, convolutional neural networks framework).
Multiple key points are pressed preset rules by critical point detection module 115, multiple key points for obtaining temperature image
The crucial point sequence of generation first.
In the present embodiment, critical point detection module 115 is used to be obtained according to the temperature figure that temperature image collection module 114 is obtained
To multiple key points including nose, neck, shoulder etc., the key point can be arranged on different portions in specifically with scene
Position, and crucial point sequence is constituted in certain sequence., it will be clear that the default rule can be set according to concrete scene
's.For example, key point can be linked in sequence by human body contour outline.
Behavior judge module 116 is fallen down, for according to the corresponding first key point sequence of continuous first image sequence of multiframe
Row judge whether human body is fallen down.
In the present embodiment, the corresponding first crucial point sequence of the first image sequence is put into key point series model, when
Continuously the change of the corresponding described first crucial point sequence of multiframe described first image sequence is more than in key point series model
During the threshold value of the change of the corresponding described second crucial point sequence of the second image sequence described in continuous multiframe, the human body is judged
Fall down, the threshold value can be according to specific environment parameter setting.In specific implementation process, the corresponding pass of continuous image is collected
Key point sequence, judges that human body attitude changes according to crucial point sequence.
When it is implemented, in the case of human body difference posture, each key point information includes key point position, each key point with
Heart point angle etc. has different patterns, for example, people is in the case where falling down, each key point of whole body and central point angle
Degree, will be with varying widely in the case of standing or walking.Further, since the behavior of people is a continuous process, it is a succession of
The change of different postures, here, taking into full account temporal information, three-dimensional is expanded to by the two-dimensional signal of script.
Alarm module 117, for when judging falling over of human body, generating warning message.
When it is implemented, the corresponding human body attitude of the corresponding key point sequence of continuous first image is gives birth to when falling down behavior
Into warning message.The warning message can be sent to and fall down the client of the connection of warning device 100, in order to point out medical care
Personnel are given treatment in time.
Referring to Fig. 4, being the high-level schematic functional block diagram provided in an embodiment of the present invention for falling down warning device 110.Further
Ground, falling down warning device 110 also includes human body estimation model training module 118.
Human body estimates model training module 118, and for obtaining human body picture, rower is entered to the profile of the human body in picture
Note, builds deep learning network 300, and human body picture is inputted into deep learning network 300 is trained, to obtain being based on depth
Practise the human body attitude estimation model of algorithm.
In the present embodiment, a large amount of pictures for carrying somatic data are obtained, the human body contour outline in the picture are labeled, structure
Deep learning network 300 is built, human body picture is inputted into deep learning network 300 is trained, and obtains being based on deep learning algorithm
Human body attitude estimation model.
Referring to Fig. 5, being the high-level schematic functional block diagram provided in an embodiment of the present invention for falling down warning device 110.Further
Ground, falling down warning device 110 also includes key point series model module 119.
Key point series model module 119, for obtaining human body corresponding second image sequence in different postures, is obtained
The corresponding second crucial point sequence of second image sequence, posture includes falling down behavior and does not fall down behavior;According to the second image sequence
Arrange corresponding posture the second crucial point sequence is marked, obtain the corresponding human body attitude change of the second key point sequence, with
Obtain key point series model.
When it is implemented, to be also trained before critical point detection is carried out to key point series model, training process is such as
Under:
Prepare data:COCO data sets are used herein, and the personage in picture is labeled.Convert data to LMDB
Form, is easy to input caffe.
Deep learning network 300 is built, and the structure of network 300 is written as the prototxt forms that caffe receives.
Network 300 is initialized using VGG-19model, and starts training.
Further, for obtaining human body corresponding second image sequence in different postures, the second image sequence is obtained
Corresponding second crucial point sequence is specifically included:
Human body attitude model video is obtained, by Video processing into the image sequence of multiframe second;Obtain multiframe continuous second
The crucial point sequence of the second of image sequence.
When it is implemented, corresponding second image sequence is that known human body attitude model is regarded during the different postures of the acquisition
Frequently, human body attitude model video is subjected to the processing generation image sequence of multiframe second, key point is carried out in the second image sequence
Mark obtains crucial point sequence, and the corresponding human body attitude change of the second key point sequence.
Further, the every one second image sequence corresponding time in record continuous second image sequence of multiframe;Obtain
The corresponding second crucial point sequence of every one second image sequence in continuous second image sequence of multiframe.
When it is implemented, needing that the time of every one second image sequence is marked, to obtain multiframe continuous second
The change of the corresponding crucial point sequence of image, the attitudes vibration of human body is judged with this.
Referring to Fig. 6, falling down the flow that alarm method is applied to fall down warning device 100 to be provided in an embodiment of the present invention
Figure.It the described method comprises the following steps:
Step S101, obtains the ambient video in pickup area.
It is to be appreciated that step S101 can be performed by video acquiring module 112.
Step S102, the image sequence of multiframe first is processed into by ambient video.
It is to be appreciated that step S102 can be performed by image segmentation module 113.
Step S103, is input to the human body attitude based on deep learning algorithm by each image sequence of frame first and estimates mould
Type, obtains the corresponding temperature image of human body in ambient video.
It is to be appreciated that step S103 can be performed by temperature image collection module 114.
Step S104, obtains multiple key points of temperature image, and multiple key points are crucial by preset rules generation first
Point sequence.
It is to be appreciated that step S104 can be performed by critical point detection module 115.
Step S105, judges whether human body falls according to the corresponding first crucial point sequence of continuous first image sequence of multiframe
.
It is to be appreciated that step S105 can be performed by falling down behavior judge module 116.
Step S106, when judging falling over of human body, generates warning message.
It is to be appreciated that step S106 can be performed by alarm module 117.
Referring to Fig. 7, setting up the step of human body attitude estimates model to be provided in an embodiment of the present invention.It should be understood that
Estimate that model training module 118 can perform step by human body.
The step includes:
Step S201, obtains human body picture, the profile of the human body in picture is labeled.
Step S202, builds deep learning network 300.
Step S203, inputs deep learning network 300 by human body picture and is trained, to obtain being based on deep learning algorithm
Human body attitude estimation model.
Fig. 8-Figure 10 is referred to, is provided in an embodiment of the present invention the step of set up key point series model.It is appreciated that
, step once can be performed by key point series model module 119.
The step includes:
Step S301, obtains human body corresponding second image sequence in different postures, obtains the second image sequence correspondence
The second crucial point sequence, posture includes falling down behavior and do not fall down behavior.
Step S302, is marked to the second crucial point sequence according to the corresponding posture of the second image sequence, obtains second
The corresponding human body attitude change of key point sequence, to obtain key point series model.
The step S301 specifically includes step S3011 and step S3012.
Step S3011, obtains human body attitude model video, by Video processing into the image sequence of multiframe second.
Step S3012, obtains the second crucial point sequence of continuous second image sequence of multiframe.
The step S3012 includes step S30121 and step S30122.
The every one second image sequence corresponding time in step S30121, record continuous second image sequence of multiframe.
Step S30122, obtains every one second image sequence in continuous second image sequence of multiframe corresponding second crucial
Point sequence.
In summary, alarm method, device and equipment are fallen down the invention provides one kind, is related to falling-resistant monitoring field.
This method is applied to the device, and the equipment includes the device, and this method includes:Obtain the ambient video in pickup area;By ring
Border Video processing is into the image sequence of multiframe first;Each image sequence of frame first is input to the human body based on deep learning algorithm
Attitude estimation model, obtains the corresponding temperature image of human body in ambient video;Multiple key points of temperature image are obtained, will be multiple
Key point is by the crucial point sequence of preset rules generation first;According to corresponding first key point of continuous first image sequence of multiframe
Sequence judges whether human body is fallen down;When judging falling over of human body, warning message is generated.By gathering the human body in ambient video,
The change of human body attitude is obtained, warning message is generated when judging falling over of human body, is worn without human body, do not influence physical activity
Body state is detected simultaneously.
, can also be by other in embodiment provided herein, it should be understood that disclosed apparatus and method
Mode realize.Device embodiment described above is only schematical, for example, the flow chart and block diagram in accompanying drawing are shown
According to the device, the architectural framework in the cards of method and computer program product, function of multiple embodiments of the present invention
And operation.At this point, each square frame in flow chart or block diagram can represent one of a module, program segment or code
Point, a part for the module, program segment or code is used to realize the executable of defined logic function comprising one or more
Instruction.It should also be noted that in some implementations as replacement, the function of being marked in square frame can also be with different from attached
The order marked in figure occurs.For example, two continuous square frames can essentially be performed substantially in parallel, they also may be used sometimes
To perform in the opposite order, this is depending on involved function.It is also noted that each in block diagram and/or flow chart
The combination of square frame and the square frame in block diagram and/or flow chart, can with function or action as defined in performing it is special based on
The system of hardware is realized, or can be realized with the combination of specialized hardware and computer instruction.
The preferred embodiments of the present invention are the foregoing is only, are not intended to limit the invention, for the skill of this area
For art personnel, the present invention can have various modifications and variations.Within the spirit and principles of the invention, that is made any repaiies
Change, equivalent substitution, improvement etc., should be included in the scope of the protection.