CN109872359A - Sitting posture detecting method, device and computer readable storage medium - Google Patents
Sitting posture detecting method, device and computer readable storage medium Download PDFInfo
- Publication number
- CN109872359A CN109872359A CN201910076996.7A CN201910076996A CN109872359A CN 109872359 A CN109872359 A CN 109872359A CN 201910076996 A CN201910076996 A CN 201910076996A CN 109872359 A CN109872359 A CN 109872359A
- Authority
- CN
- China
- Prior art keywords
- human body
- sitting posture
- body target
- video image
- model
- 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.)
- Pending
Links
Abstract
The present invention discloses a kind of sitting posture detecting method, device and computer readable storage medium.The described method includes: obtaining the video image based on visible or infrared light of selection area；Judge in the video image with the presence or absence of human body target；If there are human body targets in the video image, relative position and the arrangement of the skeleton model of human body target are detected；According to the skeleton model relative position and arrangement come judge human body target whether be sit state；If the human body target is the state sat, judge whether the sitting posture of the human body target is abnormal；If the sitting posture of the human body target is abnormal, exception information is sent to monitor terminal.The present invention has user experience good, easy to use, highly-safe, and detection efficiency is high, the high advantage of accuracy rate.
Description
Technical field
The present invention relates to sitting posture detection technique field, in particular to a kind of sitting posture detecting method based on visible light video,
Device and computer readable storage medium.
Background technique
Nowadays, due to the needs of study and work, the time of the daily sitting of people is increasingly longer, according to statistics, nowadays people
The time of average daily sitting is more than 8 hours.In prolonged sitting, due to people be difficult to be always maintained at it is good
Standing and sitting gesture, therefore it is easy to appear the incorrect sitting-poses such as partially of too close, body inclination, head from desk.Over time, it will be easy to
Suffer from myopia, strabismus, lumbar vertebral disease, cervical spondylosis etc., the serious health for influencing people.
By taking youngsters and children as an example, currently, the youngsters and children myopia disease incidence in China has occupied second place of the world, correlation money
Material display: students' rate of myopia increasingly rises, and university student is up to 70%, middle school student up to 30%~40%, pupil up to 20%.
Meanwhile student's ratio of the spinal curvature really generated by abnormal sitting posture is also very high.Prevent youngsters and children myopia and hunchback is complete
Social concern.Phenomena such as sitting posture is crooked, with one's back bent, sighting distance is excessively close when children for learning is to lead to myopia, strabismus and neck
The main reason for vertebra depauperation.The current main method for solving wrong sitting posture both at home and abroad is the rectifier of physics class, such as ear-hang
Formula sitting position corrector, spondyle appliance etc., they play pre- myopic-preventing purpose from physical angle mostly, and require user long
Phase wears these rectifiers.And long-time wear corrective device, it may result in the uncomfortable or even bored of user.Utilize ultrasonic wave
The rectifier of ranging sensing technology exploitation, plays certain effect, but blocked under normal conditions because of penholder and arm
It influences, so that systematic error rate is larger, arrive unhappy experience to user.
Therefore, in order to promote people to correct incorrect sitting-pose, form good sitting posture habit, reduce and suffer from myopia, lumbar vertebral disease
Etc. diseases probability, carry out sitting posture detection be necessary.Currently, having many sitting posture detections, pre- myopia prevention, correcting and sit
The technological means of appearance.Wherein, most common method is exactly to pass through sensor (sensors such as infrared, pressure, acceleration, ultrasonic wave)
It realizes.However, there is complicated installation, using trouble and sitting posture accuracy in detection in existing sitting posture detecting method or device
Not high disadvantage.
Summary of the invention
Present invention solves the technical problem that it is accurate to be to provide a kind of detection, sitting posture detecting method easy to use.
To achieve the above object, the present invention provides a kind of sitting posture detecting methods comprising:
Obtain the video image based on visible or infrared light of selection area；
Judge in the video image with the presence or absence of human body target；
If there are human body targets in the video image, relative position and the row of the skeleton model of human body target are detected
Column；
According to the skeleton model relative position and arrangement come judge human body target whether be sit state；
If the human body target is the state sat, judge whether the sitting posture of the human body target is abnormal；
If the sitting posture of the human body target is abnormal, exception information is sent to monitor terminal.
Preferably, whether Yi Chang method includes: the sitting posture for judging the human body target
The sitting posture human body target that will test is reduced to human body sitting posture skeleton model；
Judge whether the head model of the human body sitting posture skeleton model is overlapped with shoulder model；
If head model is overlapped with shoulder model, judge the sitting posture of the human body target for exception.
Preferably, after whether the head model for judging the human body sitting posture skeleton model is overlapped with shoulder model
Further include:
If head model is not overlapped with shoulder model, judge the difference of the ordinate on shoulder model both sides whether default
In range；
If the difference of the ordinate on shoulder model both sides within a preset range, does not judge that the sitting posture of the human body target is
It is abnormal.
Preferably, it is also wrapped whether it is within the preset range in the difference of the ordinate on the judgement shoulder model both sides
It includes:
If the difference of the ordinate on shoulder model both sides is within a preset range, judge whether human trunk model is vertical about itself
Coordinate is symmetrical；
If asymmetric, judge that the sitting posture of the human body target for exception, otherwise judges that the sitting posture of the human body target is
Normally.
Preferably, the sitting posture detecting method further include:
The video image is shown on mobile terminals；
Obtain the selection area that user inputs on the mobile terminal；
According to the video image, it is default to judge whether the frontier distance of the human body target and the selection area is less than
Value；
When the frontier distance of the human body target and the selection area is less than preset value, then indicating risk letter is issued
Breath.
Preferably, the selection area that the acquisition user inputs on the mobile terminal specifically includes:
Obtain the selection area that user encloses picture by finger on the video image of the mobile terminal.
Preferably, the selection area that the acquisition user inputs on the mobile terminal specifically includes:
It obtains user and passes through the selection area that voice is inputted in the mobile terminal.
It preferably, is by the human body based on deep learning with the presence or absence of human body target in the judgement video image
Bone detection method judges whether to have human body target in video image.
Another aspect of the present invention also provides a kind of sitting posture detection device, comprising:
Image collection module, for obtaining the video image based on visible light of selection area；
First judgment module, for judging in the video image with the presence or absence of human body target；
Detection module, if detecting the skeleton model of human body target for there are human body targets in the video image
Relative position and arrangement；
Second judgment module, for according to the skeleton model relative position and arrangement whether judge human body target
For the state of seat；
Third judgment module judges that the sitting posture of the human body target is if being the state sat for the human body target
No exception；
Exception information is sent to monitor terminal if the sitting posture for the human body target is abnormal by information sending module.
Another aspect of the present invention also provides a kind of computer readable storage medium, the computer-readable recording medium storage
There is one or more program, one or more of programs can be executed by one or more processor, as above to realize
State each step of any sitting posture detecting method.
Sitting posture detecting method, device and computer readable storage medium of the invention has the following beneficial effects: by institute
The method of stating includes the video image based on visible light for obtaining selection area；Judge in the video image with the presence or absence of human body
Target；If there are human body targets in the video image, relative position and the arrangement of the skeleton model of human body target are detected；According to
Judge whether human body target is the state sat according to the relative position and arrangement of the skeleton model；If the human body target is
The state of seat then judges whether the sitting posture of the human body target is abnormal；If the sitting posture of the human body target is abnormal, by abnormal letter
Breath is sent to monitor terminal.That is, sitting posture state is obtained by way of visible light video, without the body with human body target
Body generates contact, can detect the sitting posture state of human body target in real time, thus, user experience is good, easy to use and safety
Height, in addition, relative position and arrangement by the skeleton model of detection human body target, to determine whether human body target is the shape sat
Whether abnormal state further detects sitting posture, thus detection efficiency is high, and accuracy rate is also high.
Detailed description of the invention
Fig. 1 is the flow chart of sitting posture detecting method preferred embodiment of the present invention；
Fig. 2 be sitting posture detecting method of the present invention the sitting posture for judging the human body target whether Yi Chang flow chart；
Fig. 3 is the functional block diagram of sitting posture detection device of the present invention.
Specific embodiment
Invention is described in detail with reference to the accompanying drawings and examples.It should be noted that invention is real if do not conflicted
The each feature applied in example and embodiment can be combined with each other, within the protection scope of invention.
Embodiment 1
Fig. 1 and Fig. 2 is please referred to, the present invention provides a kind of sitting posture detecting methods comprising following steps:
S100, the video image based on visible light for obtaining selection area；
Selection area is shot based on the photographic device of visible light by CCD camera etc., to obtain human body target.
In order to which picture is more complete, the accuracy rate of detection is improved, photographic device face human body target height direction is shot.
S200, judge in the video image with the presence or absence of human body target；
It can be compared according to the video image and pre-set image taken, to judge in the video image with the presence or absence of human body
Target.Preferably, in the present embodiment, it is in video image using being judged based on the skeleton detection method of deep learning
It is no to have human body target.
The deep learning network of the present embodiment is made using visual angle geometry pre-training network (VGG pre-trainnetwork)
For skeleton, L (p) and S (p) are returned respectively there are two branch.Each stage (stage) calculates primary loss (loss), later L
With S and original input data, continue the training in next stage.With the increase of the number of iterations, S being capable of area to a certain extent
The left and right of separation structure.The reference standard (ground-truth) of the L2 norm of loss (loss), S and L are needed from the pass of mark
Key point generates, and does not calculate the point if some key point has missing in the callout.
For S, every one kind key point has a channel (channel), generate reference standard (ground-truth) when
Time is to be maximized the method for (max) according to multiple Gaussian Profiles to retain the peak value of response of each point.For L then complexity one
Point, referring initially to accurate definition, the PAFs (Part Affinity Fields) done for c-th of limb of k-th of person:
The position of xj, k expression k-th of person, j-th of key point.And whether pixel p falls in limb (limb is dry) and then sets
One threshold range:
Wherein lc, k and σ l respectively indicate limb length and width.Finally all person the same category limbs can also be done into
Row is average, so that the channel (channel) of the output of L is equal with dry kind of number of limb:
After knowing PAFs (Part Affinity Fields) and key point position dj, need to assess this two key point
Correlation.Then on the two key point lines the dot product of each pixel PAF vector and line vector integral:
After obtaining key node and side right, calculates posture skeleton and be fully converted into a figure problem.So
Afterwards, using Hungary Algorithm to adjacent node carried out Optimum Matching (such as a pile left finesse node and a pile left hand toggle point,
Side right is calculated using the PAF of forearm, then carries out Optimum Matching), finally obtain entire humanoid posture skeleton.The present embodiment is based on
The skeleton detection method of deep learning judges whether to have in video image human body target, thus has the detection speed fast,
High-efficient and high accuracy rate advantage.It is connected between artis with line, head and shoulder.Artis is automatically determined according to manikin, it will
Artis connection building virtual skeleton model (being obtained for example, by using open pose algorithm).The present invention is judged by image recognition
It determines the sitting posture of people, the sitting posture of people is then calculated using the coordinate of normal projection.
If there are human body targets in S300, the video image, the relative position of the skeleton model of human body target is detected
And arrangement；
Wherein, the skeleton model includes the skeleton model of head, trunk and four limbs, detects the skeleton model of human body target
Relative position and arrangement can relative position for head and trunk and placement relationship, of course, it is possible to be examined as needed
Gauge head portion, the relative position of trunk and four limbs and placement relationship.
S400, the relative position according to the skeleton model and arrangement come judge human body target whether be sit state；
For example, being judged as seat if head is equal at a distance from a certain section of each region of thigh or is less than preset value
State.If certain hand is equal at a distance from a certain section of each region of thigh or is less than preset value, it is judged as the shape of seat
State.It is judged as the state of seat if knee is higher than finger.
If S500, the human body target are the state sat, judge whether the sitting posture of the human body target is abnormal；
For example, if head is unequal at a distance from two shoulders or greater than preset value, judges that the sitting posture of the human body target is
It is abnormal.If the angle that face and chest are formed is less than preset value, judge the sitting posture of the human body target for exception.
If the sitting posture of S600, the human body target is abnormal, exception information is sent to monitor terminal.
The monitor terminal can be mobile phone, laptop or remote monitor etc., be not specifically limited herein,
As long as it can receive the exception information.Application program for receiving and sending messages is installed on the monitor terminal
(app), user can be by the monitor terminal sitting posture of monitoring objective, thus uses more convenient.
In the preferred embodiment of the present invention, the sitting posture for judging the human body target whether Yi Chang method packet
It includes:
S101, the sitting posture human body target that will test are reduced to human body sitting posture skeleton model；
S102, judge whether the head model of the human body sitting posture skeleton model is overlapped with shoulder model；
If S103, head model are overlapped with shoulder model, judge the sitting posture of the human body target for exception.
The advantage that detection efficiency is high and accuracy rate is high can be realized by the detection method.
In the preferred embodiment of the present invention, the head model for judging the human body sitting posture skeleton model whether
After being overlapped with shoulder model further include:
If head model is not overlapped with shoulder model, judge the difference of the ordinate on shoulder model both sides whether default
In range；
If the difference of the ordinate on shoulder model both sides within a preset range, does not judge that the sitting posture of the human body target is
It is abnormal.
In the preferred embodiment of the present invention, the ordinate for judging shoulder model both sides difference whether pre-
If after in range further include:
If the difference of the ordinate on shoulder model both sides is within a preset range, judge whether human trunk model is vertical about itself
Coordinate is symmetrical；
If asymmetric, judge that the sitting posture of the human body target for exception, otherwise judges that the sitting posture of the human body target is
Normally.
By above-mentioned detecting step, thus the accuracy rate of judgement can be improved.
In a preferred embodiment of the invention, the sitting posture detecting method further include:
The video image is shown on mobile terminals；
Obtain the selection area that user inputs on the mobile terminal；
According to the video image, it is default to judge whether the frontier distance of the human body target and the selection area is less than
Value；
When the frontier distance of the human body target and the selection area is less than preset value, then indicating risk letter is issued
Breath.
Through the above steps, abnormal special population can detects to body, thus not only accuracy rate is high, more just
It is used in user, and application is more extensive.
In the preferred embodiment of the present invention, the selection area tool for obtaining user and inputting on the mobile terminal
Body includes: the selection area for obtaining user and enclosing picture on the video image of the mobile terminal by finger.
That is user on the video image that the display screen in mobile phone is shown by touch by way of circle draw one
The region is defined as selection area by region.Wherein, the display screen is touch screen.This kind input selection area mode compared with
User-friendly, user experience is good, and monitoring efficiency is high.
In the preferred embodiment of the present invention, the selection area tool for obtaining user and inputting on the mobile terminal
Body includes: to obtain user to pass through the selection area that voice is inputted in the mobile terminal.
That is user is inputted by way of voice, then the mobile terminals such as mobile phone are just automatically according to the language of acquisition
Sound signal is configured the selection area.Such mode preferably avoids false triggering, so that the problem of mistaking instruction.
In the preferred embodiment of the present invention, it whether there is the step of human body target in the judgement video image
In rapid further include: detect whether that there are human face targets.It can be monitored detection to specific people by the step, and can mention
The accuracy rate of high detection.
It is described detect whether there are human face target to be calculated by convolutional neural networks in the preferred embodiment of the present invention
Method realizes the video image processing.By convolutional neural networks algorithm to the video image processing, thus preferably keep away
Exempt to lead to the problem for identifying accuracy difference posture, illumination or due to blocking etc..
In the preferred embodiment of the present invention, the convolutional neural networks algorithm includes the following steps:
Candidate forms are quickly generated to the video image processing by the first convolutional neural networks；
In the preferred embodiment of the present invention, first convolutional neural networks that pass through quickly generate candidate forms packet
It includes: using full convolutional neural networks to the video image processing, to obtain candidate forms and boundary regression vector, meanwhile, it waits
It selects forms to be calibrated according to bounding box, then utilizes non-maxima suppression method removal overlapping forms.
The candidate forms are refined by the second convolutional neural networks, abandon least partially overlapped forms, wherein described the
The convolution number of plies of two convolutional neural networks is greater than the convolution number of plies of first convolutional neural networks；
It is described to pass through the second convolutional neural networks refining candidate forms in the preferred embodiment of the present invention, it loses
Abandoning least partially overlapped forms includes: to refine the picture comprising the candidate forms in third convolutional neural networks, the network
It selects the mode connected entirely to be trained, finely tunes candidate forms using bounding box vector, recycle the removal of non-maxima suppression method
It is overlapped forms.
The candidate forms are refined by third convolutional neural networks, while showing the facial characteristics point of preset quantity
It sets, wherein the convolution number of plies of the third convolutional neural networks is greater than the convolution number of plies of second convolutional neural networks.
Specifically, in the present embodiment, original image to be detected generates various sizes of image after changing size,
And construct input of the image pyramid as network.The image pyramid of building, the number of plies are determined that first is by two factors
The minimum face minSize of setting, second is zoom factor factor, minimum face representation min (w, h), in the present embodiment
In, minimum face cannot be less than 12, zoom factor 0.709, the number of plies of image pyramid can be calculated according to formula:
MinL=org_L* (12/minsize) * factor^ (n), n={ 0,1,2,3 ..., N }；
Wherein n is exactly the pyramidal number of plies, and org_L is the minimum edge min (W, H) for inputting original image, and minisize is
It is artificially set according to application scenarios, in the case where guaranteeing that minL is greater than 12, all n just constitute pyramidal layer.So
The value of minsize is smaller, and the value range of n is bigger, and calculation amount is correspondingly increased, and the face being able to detect that is smaller.
First stage quickly generates candidate window by the convolutional neural networks of shallow-layer, and the network is all by convolutional layer reality
It is existing, the regression vector of candidate face window and face window is got, the regression vector based on face window is corrected face window, so
Non-maxima suppression (NMS) is carried out to all face windows afterwards, merges the face window of high superposed.Its detailed process is exactly to pass through figure
As the picture for the various sizes size that pyramid generates, each figure all carries out a propagated forward, obtains on each figure
As a result remove a part using the threshold value of setting after, it is remaining that coordinate in original image is reverted to according to zoom scale, will own
Coordinate information summarize, then carry out non-maxima suppression and remove a part of redundancy.
Second stage is mistaken as face by a more complicated convolutional neural networks again to handle in the first stage
" face window " to refine face window, input of the output of first stage as second stage, the first stage is finally produced
They after pushing back these bounding boxes in original image according to zoom factor, are all modified size and arrived by a large amount of bounding box
24x24 size, the input as second stage.Second stage is by equally generating a large amount of bounding box after network, similarly
Remove a part according to threshold value, non-maxima suppression method is recycled to remove a part.
Finally using the bounding box finally stayed in second stage, after reverting on original picture, all modify
Size is then input to the phase III to 48x48 size, further refines result by the convolutional neural networks of phase III
And export 5 characteristic points on face.Processing from coarse to fine is carried out to task by the concatenated convolutional neural network of three ranks,
Final output face frame position and five characteristic point positions are cascaded using correlation intrinsic between detection and calibration in depth
Multitask frame get off to be promoted their performance, have the advantages that precision height and real-time are good.
In the preferred embodiment of the present invention, it is described detect whether there are human face target before further include walking as follows
It is rapid:
Read in facial image；
It asks for help face sample mean face；
Obtain non-face feature；
Face and non-face differentiation are carried out by linear classifier, and exports differentiation result.
That is, train face and non-face sample first before formally carrying out business use, with obtain face with
Non-face feature, the result for recycling training to obtain is for detecting.Wherein, average face is averaging to overall face sample,
When carrying out processing and feature extraction to image after reading image, average face is first subtracted.Due to carrying out face by linear classifier
Classify with non-face two classifications, thus keep the degree of polymerization in the classes of two classes fine, thus can be improved differentiate face with it is inhuman
The accuracy rate of face.
From the foregoing, it will be observed that since the method includes obtaining the video image based on visible light of selection area；Described in judgement
It whether there is human body target in video image；If there are human body targets in the video image, the bone of human body target is detected
The relative position of model and arrangement；Judge whether human body target is seat according to the relative position of the skeleton model and arrangement
State；If the human body target is the state sat, judge whether the sitting posture of the human body target is abnormal；If the human body mesh
Target sitting posture is abnormal, then exception information is sent to monitor terminal.That is, obtaining sitting posture by way of visible light video
State contacts without generating with the body of human body target, can detect the sitting posture state of human body target in real time, thus, user
It experiences, it is easy to use and highly-safe, in addition, relative position and arrangement by the skeleton model of detection human body target, come
Determine whether human body target is the state sat, and whether abnormal further detects sitting posture, thus detection efficiency is high, accuracy rate
It is high.
Embodiment 2
Referring to Fig. 3, the present invention also provides a kind of sitting posture detection devices, comprising:
Image collection module 1, for obtaining the video image based on visible light of selection area；
First judgment module 2, for judging in the video image with the presence or absence of human body target；
Detection module 3, if detecting the skeleton model of human body target for there are human body targets in the video image
Relative position and arrangement；
Second judgment module 4, for according to the skeleton model relative position and arrangement to judge human body target be
The no state to sit；
Third judgment module 5 judges that the sitting posture of the human body target is if being the state sat for the human body target
No exception；
Exception information is sent to monitoring eventually if the sitting posture for the human body target is abnormal by information sending module 6
End.
Since the detection method of the sitting posture detection device includes the video figure based on visible light for obtaining selection area
Picture；Judge in the video image with the presence or absence of human body target；If there are human body targets in the video image, human body is detected
The relative position of the skeleton model of target and arrangement；Human body mesh is judged according to the relative position of the skeleton model and arrangement
Whether mark is the state sat；If the human body target is the state sat, judge whether the sitting posture of the human body target is abnormal；If
The sitting posture of the human body target is abnormal, then exception information is sent to monitor terminal.That is, the side for passing through visible light video
Formula obtains sitting posture state, contacts without generating with the body of human body target, can detect the sitting posture state of human body target in real time,
Thus, user experience is good, and it is easy to use and highly-safe, in addition, the relative position of the skeleton model by detection human body target
And whether abnormal arrangement further detects sitting posture to determine whether human body target is the state sat, thus detection efficiency is high,
Accuracy rate is also high.
Embodiment 3
Another aspect of the present invention also provides a kind of computer readable storage medium, the computer-readable recording medium storage
There is one or more program, one or more of programs can be executed by one or more processor, following to realize
Step:
Obtain the video image based on visible light of selection area；
Judge in the video image with the presence or absence of human body target；
If there are human body targets in the video image, relative position and the row of the skeleton model of human body target are detected
Column；
According to the skeleton model relative position and arrangement come judge human body target whether be sit state；
If the human body target is the state sat, judge whether the sitting posture of the human body target is abnormal；
If the sitting posture of the human body target is abnormal, exception information is sent to monitor terminal.
In a preferred embodiment, the whether abnormal selection area step of the sitting posture for judging the human body target
In rapid, one or more of programs can be executed by one or more processor, to perform the steps of
The sitting posture human body target that will test is reduced to human body sitting posture skeleton model；
Judge whether the head model of the human body sitting posture skeleton model is overlapped with shoulder model；
If head model is overlapped with shoulder model, judge the sitting posture of the human body target for exception.
In a preferred embodiment, the head model for judging the human body sitting posture skeleton model whether with
In the step of after shoulder model coincidence, one or more of programs can be executed by one or more processor, with reality
Existing following steps:
If head model is not overlapped with shoulder model, judge the difference of the ordinate on shoulder model both sides whether default
In range；
If the difference of the ordinate on shoulder model both sides within a preset range, does not judge that the sitting posture of the human body target is
It is abnormal.
In a preferred embodiment, in the difference of the ordinate for judging shoulder model both sides whether default
In the step of after in range, one or more of programs can be executed by one or more processor, following to realize
Step:
If the difference of the ordinate on shoulder model both sides is within a preset range, judge whether human trunk model is vertical about itself
Coordinate is symmetrical；
If asymmetric, judge that the sitting posture of the human body target for exception, otherwise judges that the sitting posture of the human body target is
Normally.
In a preferred embodiment, one or more of programs can be held by one or more processor
Row, to perform the steps of
The video image is shown on the mobile terminal；
Obtain the selection area that user inputs on the mobile terminal；
According to the video image, it is default to judge whether the frontier distance of the human body target and the selection area is less than
Value；
When the frontier distance of the human body target and the selection area is less than preset value, then indicating risk letter is issued
Breath.
In a preferred embodiment, the selection area step for obtaining user and inputting on the mobile terminal
In, one or more of programs can be executed by one or more processor, to perform the steps of
Obtain the selection area that user encloses picture by finger on the video image of the mobile terminal.
In a preferred embodiment, the selection area step for obtaining user and inputting on the mobile terminal
In, one or more of programs can be executed by one or more processor, to perform the steps of
It obtains user and passes through the selection area that voice is inputted in the mobile terminal.
In a preferred embodiment, described to detect whether in the step of there are human face targets to be to pass through convolutional Neural
Network algorithm realizes the video image processing.
Detailed Jie has been carried out to sitting posture detecting method provided by invention, device and computer readable storage medium above
It continues, specific examples are used herein to describe the principles and implementation manners of the present invention, and the explanation of above embodiments is
It is used to help understand the method and its core concept of invention；At the same time, for those skilled in the art, the think of according to invention
Think, there will be changes in the specific implementation manner and application range.In conclusion the content of the present specification is only the reality invented
Mode is applied, the scope of the patents of invention is not intended to limit, it is all to utilize equivalent structure made by description of the invention and accompanying drawing content
Or equivalent process transformation, be applied directly or indirectly in other relevant technical fields, similarly include invention patent protect
It protects in range, should not be construed as the limitation to invention.
Claims (10)
1. a kind of sitting posture detecting method, which comprises the steps of:
Obtain the video image based on visible or infrared light of selection area；
Judge in the video image with the presence or absence of human body target；
If there are human body targets in the video image, relative position and the arrangement of the skeleton model of human body target are detected；
According to the skeleton model relative position and arrangement come judge human body target whether be sit state；
If the human body target is the state sat, judge whether the sitting posture of the human body target is abnormal；
If the sitting posture of the human body target is abnormal, exception information is sent to monitor terminal.
2. sitting posture detecting method as described in claim 1, which is characterized in that whether the sitting posture for judging the human body target
Abnormal method includes:
The sitting posture human body target that will test is reduced to human body sitting posture skeleton model；
Judge whether the head model of the human body sitting posture skeleton model is overlapped with shoulder model；
If head model is overlapped with shoulder model, judge the sitting posture of the human body target for exception.
3. sitting posture detecting method as claimed in claim 2, which is characterized in that in the judgement human body sitting posture skeleton model
Head model whether be overlapped with shoulder model after further include:
If head model is not overlapped with shoulder model, judge the difference of the ordinate on shoulder model both sides whether in preset range
It is interior；
If the difference of the ordinate on shoulder model both sides not within a preset range, judges the sitting posture of the human body target to be different
Often.
4. sitting posture detecting method as claimed in claim 3, which is characterized in that the ordinate on the judgement shoulder model both sides
Difference whether it is within the preset range further include:
If the difference of the ordinate on shoulder model both sides is within a preset range, judge human trunk model whether about itself ordinate
Symmetrically；
If asymmetric, judge that the sitting posture of the human body target for exception, otherwise judges that the sitting posture of the human body target is normal.
5. such as the described in any item sitting posture detecting methods of Claims 1-4, which is characterized in that the sitting posture detecting method also wraps
It includes:
The video image is shown on mobile terminals；
Obtain the selection area that user inputs on the mobile terminal；
According to the video image, judge whether the human body target and the frontier distance of the selection area are less than preset value；
When the frontier distance of the human body target and the selection area is less than preset value, then indicating risk information is issued.
6. sitting posture detecting method as claimed in claim 5, which is characterized in that the acquisition user is defeated on the mobile terminal
The selection area entered specifically includes:
Obtain the selection area that user encloses picture by finger on the video image of the mobile terminal.
7. sitting posture detecting method as claimed in claim 5, which is characterized in that the acquisition user is defeated on the mobile terminal
The selection area entered specifically includes:
It obtains user and passes through the selection area that voice is inputted in the mobile terminal.
8. such as the described in any item sitting posture detecting methods of Claims 1-4, which is characterized in that the judgement video image
In with the presence or absence of human body target be by judging whether have in video image based on the skeleton detection method of deep learning
Human body target.
9. a kind of sitting posture detection device characterized by comprising
Image collection module, for obtaining the video image based on visible light of selection area；
First judgment module, for judging in the video image with the presence or absence of human body target；
Detection module, if detecting the opposite of the skeleton model of human body target for there are human body targets in the video image
Position and arrangement；
Second judgment module, for according to the skeleton model relative position and arrangement judge whether human body target is seat
State；
Third judgment module judges whether the sitting posture of the human body target is different if being the state sat for the human body target
Often；
Exception information is sent to monitor terminal if the sitting posture for the human body target is abnormal by information sending module.
10. a kind of computer readable storage medium, which is characterized in that the computer-readable recording medium storage have one or
Multiple programs, one or more of programs can be executed by one or more processor, to realize that claim 1-8 such as appoints
Each step of sitting posture detecting method described in one.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910076996.7A CN109872359A (en) | 2019-01-27 | 2019-01-27 | Sitting posture detecting method, device and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910076996.7A CN109872359A (en) | 2019-01-27 | 2019-01-27 | Sitting posture detecting method, device and computer readable storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109872359A true CN109872359A (en) | 2019-06-11 |
Family
ID=66918097
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910076996.7A Pending CN109872359A (en) | 2019-01-27 | 2019-01-27 | Sitting posture detecting method, device and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109872359A (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110210446A (en) * | 2019-06-12 | 2019-09-06 | 广东工业大学 | A kind of sitting posture condition detection method, device, equipment and the medium of target object |
| CN110443147A (en) * | 2019-07-10 | 2019-11-12 | 广州市讯码通讯科技有限公司 | A kind of sitting posture recognition methods, system and storage medium |
| CN110503083A (en) * | 2019-08-30 | 2019-11-26 | 北京妙医佳健康科技集团有限公司 | A kind of critical point detection method, apparatus and electronic equipment |
| CN110934591A (en) * | 2019-09-30 | 2020-03-31 | 宁波华茂优加科技有限公司 | Sitting posture detection method and device |
| CN111091046A (en) * | 2019-10-28 | 2020-05-01 | 北京灵鹦科技有限公司 | User bad sitting posture correction system based on machine vision |
| CN111199211A (en) * | 2019-12-31 | 2020-05-26 | 武汉星巡智能科技有限公司 | Intelligent monitoring equipment with infrared awakening function, monitoring method and storage medium |
| CN111265220A (en) * | 2020-01-21 | 2020-06-12 | 王力安防科技股份有限公司 | Myopia early warning method, device and equipment |
| CN111402987A (en) * | 2019-12-31 | 2020-07-10 | 武汉星巡智能科技有限公司 | Medication reminding method, device and equipment based on visible light video and storage medium |
| CN111860183A (en) * | 2020-06-23 | 2020-10-30 | 佛山电器照明股份有限公司 | Desk lamp system with sitting posture correction function and control method thereof |
| CN112200088A (en) * | 2020-10-10 | 2021-01-08 | 普联技术有限公司 | Sitting posture monitoring method, device, equipment and system |
| CN113051991A (en) * | 2020-11-08 | 2021-06-29 | 泰州市出彩网络科技有限公司 | Abnormal state dual detection platform and method |
| CN113312938A (en) * | 2020-02-26 | 2021-08-27 | 北京君正集成电路股份有限公司 | Method and system for preventing false alarm generated when no target exists in front of detector in sitting posture detection |
| CN113313917A (en) * | 2020-02-26 | 2021-08-27 | 北京君正集成电路股份有限公司 | Method for solving false alarm generated when no target exists in front of detector in sitting posture detection |
| CN113378762A (en) * | 2021-06-25 | 2021-09-10 | 宁波星巡智能科技有限公司 | Sitting posture intelligent monitoring method, device, equipment and storage medium |
| CN113780220A (en) * | 2021-09-17 | 2021-12-10 | 东胜神州旅游管理有限公司 | Child sitting posture detection method and system based on child face recognition |
| CN114038016A (en) * | 2021-11-16 | 2022-02-11 | 平安普惠企业管理有限公司 | Sitting posture detection method, device, equipment and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7366645B2 (en) * | 2002-05-06 | 2008-04-29 | Jezekiel Ben-Arie | Method of recognition of human motion, vector sequences and speech |
| CN103908063A (en) * | 2014-04-03 | 2014-07-09 | 安徽海聚信息科技有限责任公司 | Intelligent desk with sitting posture correcting function and correcting method implemented by intelligent desk |
| CN103908064A (en) * | 2014-04-03 | 2014-07-09 | 安徽海聚信息科技有限责任公司 | Intelligent desk with sitting posture correcting function and correcting method implemented by intelligent desk |
| CN106973326A (en) * | 2017-04-28 | 2017-07-21 | 张青 | It is capable of the intelligent TV set and its control system and control method of intelligent standby |
| CN107358213A (en) * | 2017-07-20 | 2017-11-17 | 湖南科乐坊教育科技股份有限公司 | A kind of children's reading is accustomed to detection method and device |
| CN107871102A (en) * | 2016-09-23 | 2018-04-03 | 北京眼神科技有限公司 | A kind of method for detecting human face and device |
-
2019
- 2019-01-27 CN CN201910076996.7A patent/CN109872359A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7366645B2 (en) * | 2002-05-06 | 2008-04-29 | Jezekiel Ben-Arie | Method of recognition of human motion, vector sequences and speech |
| CN103908063A (en) * | 2014-04-03 | 2014-07-09 | 安徽海聚信息科技有限责任公司 | Intelligent desk with sitting posture correcting function and correcting method implemented by intelligent desk |
| CN103908064A (en) * | 2014-04-03 | 2014-07-09 | 安徽海聚信息科技有限责任公司 | Intelligent desk with sitting posture correcting function and correcting method implemented by intelligent desk |
| CN107871102A (en) * | 2016-09-23 | 2018-04-03 | 北京眼神科技有限公司 | A kind of method for detecting human face and device |
| CN106973326A (en) * | 2017-04-28 | 2017-07-21 | 张青 | It is capable of the intelligent TV set and its control system and control method of intelligent standby |
| CN107358213A (en) * | 2017-07-20 | 2017-11-17 | 湖南科乐坊教育科技股份有限公司 | A kind of children's reading is accustomed to detection method and device |
Non-Patent Citations (1)
| Title |
|---|
| 余飞 等: "多级联卷积神经网络人脸检测", 《五邑大学学报（自然科学版）》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110210446A (en) * | 2019-06-12 | 2019-09-06 | 广东工业大学 | A kind of sitting posture condition detection method, device, equipment and the medium of target object |
| CN110443147A (en) * | 2019-07-10 | 2019-11-12 | 广州市讯码通讯科技有限公司 | A kind of sitting posture recognition methods, system and storage medium |
| CN110503083A (en) * | 2019-08-30 | 2019-11-26 | 北京妙医佳健康科技集团有限公司 | A kind of critical point detection method, apparatus and electronic equipment |
| CN110934591A (en) * | 2019-09-30 | 2020-03-31 | 宁波华茂优加科技有限公司 | Sitting posture detection method and device |
| CN111091046A (en) * | 2019-10-28 | 2020-05-01 | 北京灵鹦科技有限公司 | User bad sitting posture correction system based on machine vision |
| CN111199211A (en) * | 2019-12-31 | 2020-05-26 | 武汉星巡智能科技有限公司 | Intelligent monitoring equipment with infrared awakening function, monitoring method and storage medium |
| CN111402987A (en) * | 2019-12-31 | 2020-07-10 | 武汉星巡智能科技有限公司 | Medication reminding method, device and equipment based on visible light video and storage medium |
| CN111265220A (en) * | 2020-01-21 | 2020-06-12 | 王力安防科技股份有限公司 | Myopia early warning method, device and equipment |
| CN113313917A (en) * | 2020-02-26 | 2021-08-27 | 北京君正集成电路股份有限公司 | Method for solving false alarm generated when no target exists in front of detector in sitting posture detection |
| CN113312938A (en) * | 2020-02-26 | 2021-08-27 | 北京君正集成电路股份有限公司 | Method and system for preventing false alarm generated when no target exists in front of detector in sitting posture detection |
| CN113313917B (en) * | 2020-02-26 | 2022-12-16 | 北京君正集成电路股份有限公司 | Method for solving false alarm generated when no target exists in front of detector in sitting posture detection |
| CN111860183A (en) * | 2020-06-23 | 2020-10-30 | 佛山电器照明股份有限公司 | Desk lamp system with sitting posture correction function and control method thereof |
| CN112200088A (en) * | 2020-10-10 | 2021-01-08 | 普联技术有限公司 | Sitting posture monitoring method, device, equipment and system |
| CN113051991A (en) * | 2020-11-08 | 2021-06-29 | 泰州市出彩网络科技有限公司 | Abnormal state dual detection platform and method |
| CN113378762A (en) * | 2021-06-25 | 2021-09-10 | 宁波星巡智能科技有限公司 | Sitting posture intelligent monitoring method, device, equipment and storage medium |
| CN113378762B (en) * | 2021-06-25 | 2023-08-08 | 宁波星巡智能科技有限公司 | Sitting posture intelligent monitoring method, device, equipment and storage medium |
| CN113780220A (en) * | 2021-09-17 | 2021-12-10 | 东胜神州旅游管理有限公司 | Child sitting posture detection method and system based on child face recognition |
| CN114038016A (en) * | 2021-11-16 | 2022-02-11 | 平安普惠企业管理有限公司 | Sitting posture detection method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109872359A (en) | Sitting posture detecting method, device and computer readable storage medium | |
| CN104199544B (en) | Advertisement orientation put-on method based on eye tracking | |
| Le et al. | Human posture recognition using human skeleton provided by Kinect | |
| US6246779B1 (en) | Gaze position detection apparatus and method | |
| Hsieh et al. | A real time hand gesture recognition system using motion history image | |
| US20190129501A1 (en) | Interactive Motion-Based Eye Tracking Calibration | |
| KR101901591B1 (en) | Face recognition apparatus and control method for the same | |
| CN109840493A (en) | Infantal sleeping condition detection method, device and computer readable storage medium | |
| CN108304819B (en) | Gesture recognition system and method, and storage medium | |
| CN107103309A (en) | A kind of sitting posture of student detection and correcting system based on image recognition | |
| US20120087543A1 (en) | Image-based hand detection apparatus and method | |
| CN105912126B (en) | A kind of gesture motion is mapped to the adaptive adjusting gain method at interface | |
| CN106295549A (en) | Multi-orientation Face collecting method and device | |
| CN109886137A (en) | Infant sleeping posture detection method, device and computer readable storage medium | |
| CN109063625A (en) | A kind of face critical point detection method based on cascade deep network | |
| US20230237694A1 (en) | Method and system for detecting children's sitting posture based on face recognition of children | |
| Appenrodt et al. | Multi stereo camera data fusion for fingertip detection in gesture recognition systems | |
| JP6992900B2 (en) | Information processing equipment, control methods, and programs | |
| Gallo et al. | View-independent hand posture recognition from single depth images using PCA and Flusser moments | |
| WO2022074886A1 (en) | Posture detection device, posture detection method, and sleeping posture determination method | |
| Jian-Nan et al. | Key techniques of eye gaze tracking based on pupil corneal reflection | |
| JP7439932B2 (en) | Information processing system, data storage device, data generation device, information processing method, data storage method, data generation method, recording medium, and database | |
| KR101703690B1 (en) | Apparatus for recognizing iris and operating method thereof | |
| US20230245342A1 (en) | Image selection apparatus, image selection method, and non-transitory computer-readable medium | |
| CN113989832A (en) | Gesture recognition method and device, terminal equipment and storage medium |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190611 |