CN110443150A - A kind of fall detection method, device, storage medium - Google Patents
A kind of fall detection method, device, storage medium Download PDFInfo
- Publication number
- CN110443150A CN110443150A CN201910621095.1A CN201910621095A CN110443150A CN 110443150 A CN110443150 A CN 110443150A CN 201910621095 A CN201910621095 A CN 201910621095A CN 110443150 A CN110443150 A CN 110443150A
- Authority
- CN
- China
- Prior art keywords
- human body
- image
- convolutional neural
- neural networks
- fall detection
- 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
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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
This application discloses a kind of fall detection method, device, storage mediums.Wherein, fall detection method includes: that the human body image region including human body target is determined in image to be detected using the human testing model based on convolutional neural networks;And determine whether the human body target is in a falling state according to human body image region using the fall detection model based on convolutional neural networks.To which the technical solution of the application can obtain video data simultaneously from multiplex image acquisition equipment, and the detection model based on convolutional neural networks can be used and show that testing result, accuracy rate are high in real time.And since this programme uses the detection model based on convolutional neural networks, sensor is not needed relative to traditional detection method, easy to use, calculating occupancy resource is not high, and cost is relatively low, is easy to universal.
Description
Technical field
This application involves behavioral value method fields, more particularly to a kind of fall detection method, device, storage medium.
Background technique
As human longevity extends, birthrate of population is reduced, and aging has become global problem, and aging is likely to become
The normality of future world.China also strides forward towards depth aging society, the variation of young man's operating pressure and life style, nothing
Method extracts the plenty of time out and accompanies treatment old man, and the distribution of old solitary people is more and more.Due to the degeneration of old man's physiological function, fall
As the big obstacle that senior health and fitness ensures, tumble may generate great actual bodily harm to old man, if cannot get and
When give treatment to, more serious consequence may be generated.Therefore the fall detection method for studying a set of real-time high-efficiency is very necessary, thus
The old man that can help to fall after Falls Among Old People can be succoured in time, ensure the healthy living of old man.
Tumble detection method for human body and main problem are currently on the market: 1) sensor-based detection needs are carried
Wearable sensor, such detection device higher cost, is easily lost, and uses inconvenient, and area coverage is small, market is general
And application cost is also higher;2) based on the detection method of video image analysis, such method is obtained by video background modeling technique
Moving target foreground blocks are taken, are fallen by the prospect block feature of extraction, since there are many Moving Objects, such as various vehicles, respectively
Whether kind of animal, being unable to judge accurately is that people passes through, and the more accuracy rate of erroneous detection is low is also unable to reach commercial standard (CS) for detection;3) it is based on
The detection method consumption resource of deep learning model analysis picture video is larger, is delayed higher, cannot accomplish reality in video streaming
When detect, can not realize multi-path camera while detect operation, at the same can not also be transplanted to mobile terminal operation, cannot allow and fall
Old man is quickly succoured, and is extremely difficult to commercial standard (CS).
For detection accuracy present in above-mentioned existing tumble detection method for human body not high, inconvenient for use, cost compared with
Technical problem that is high, calculating higher, the universal difficulty of occupancy resource, currently no effective solution has been proposed.
Summary of the invention
Embodiment of the disclosure provides a kind of fall detection method, device and image capture device, at least to solve
Detection accuracy present in existing tumble detection method for human body is high, inconvenient for use, higher cost, calculates that occupy resource inclined
Technical problem that is high, popularizing difficulty
According to the one aspect of the embodiment of the present disclosure, a kind of fall detection method is provided, comprising: using based on convolution mind
Human testing model through network determines the human body image region including human body target in image to be detected;And utilize base
Determine whether the human body target is in tumble shape according to human body image region in the fall detection model of convolutional neural networks
State.
Optionally, using the human testing model based on convolutional neural networks, determine to include human body from image to be detected
The operation in the human body image region of target, comprising: utilize human testing model, according to image to be detected, generate respectively with it is to be checked
The corresponding multiple vectors of multiple rectangle frame regions in altimetric image, wherein vector includes at least following information: corresponding rectangle frame
It include the confidence of human body target in the location information in region, the dimension information of corresponding rectangle frame region and corresponding rectangle frame
Spend information;And rectangle frame region corresponding to the maximum vector of confidence information is determined as human body image region.
Optionally, the operation of multiple vectors corresponding with multiple rectangle frame regions in image to be detected respectively, packet are generated
It includes: multiple eigenmatrixes being generated according to image to be detected using human testing model;And using in multiple eigenmatrixes
The element of the same position of at least part eigenmatrix constructs a vector in multiple vectors.
Optionally, it using the element of the same position of at least part eigenmatrix in multiple eigenmatrixes, constructs more
The operation of a vector in a vector, comprising: multiple eigenmatrixes are divided into multiple set of matrices, wherein each matrix stack
The quantity for the eigenmatrix for including in conjunction is identical;And the member of the same position using the eigenmatrix in the same set of matrices
Element constructs a vector in multiple vectors.
Optionally, human body mesh is determined according to human body image region using the fall detection model based on convolutional neural networks
Mark whether operation in a falling state, comprising: utilize the fall detection model based on convolutional neural networks, according to human body image
Region generates the fractional value whether fallen for identifying human body target;And when fractional value is greater than predetermined threshold, determine human body
It is in a falling state.
Optionally, it is generated according to human body image region for marking using the fall detection model based on convolutional neural networks
Know the operation for the fractional value whether human body target falls, comprising: using the neural network structure in fall detection model, extract people
Characteristics of image in body image-region;And using the taxonomic structure in fall detection model, according to extracted characteristics of image,
Generate the fractional value whether fallen for identifying the human body target.
Optionally, it is generated according to human body image region for marking using the fall detection model based on convolutional neural networks
The operation for knowing the fractional value whether human body target falls, further includes the fractional value for converting fractional value to Probability Forms.
According to the other side of the embodiment of the present disclosure, a kind of falling detection device is additionally provided, comprising: human body image area
Domain determining module determines to include human body for utilizing the human testing model based on convolutional neural networks in image to be detected
The human body image region of target;And tumble state determination module, for utilizing the fall detection mould based on convolutional neural networks
Type determines whether human body target is in a falling state according to human body image region.
According to the other side of the embodiment of the present disclosure, a kind of falling detection device is additionally provided, comprising: processor;With
And memory, it is connect with processor, for providing the instruction for handling following processing step for processor: using based on convolutional Neural
The human testing model of network determines the human body image region including human body target in image to be detected;And it utilizes and is based on
The fall detection model of convolutional neural networks determines whether human body target is in a falling state according to human body image region.
To use the human testing based on convolutional neural networks using processor according to the technical solution of the present embodiment
Model analyzes image, determines the human body image region comprising human body target, and use falling based on convolutional neural networks
Detection model determines whether above-mentioned human body target is in a falling state according to the human body image region detected.The application's
Technical solution can obtain video data simultaneously from multiplex image acquisition equipment, and can be used based on convolutional neural networks
Detection model show that testing result, accuracy rate are high in real time.And since this programme uses the detection based on convolutional neural networks
Model, therefore do not need user relative to traditional detection method and carry wearable sensor, it is easy to use, it calculates and occupies resource
Not high, cost is relatively low, is easy to universal.
Therefore, the technical solution of the present embodiment solves detection accuracy present in existing tumble detection method for human body not
High, inconvenient for use, higher cost calculates the technical problem for occupying higher, the universal difficulty of resource.
Detailed description of the invention
Attached drawing described herein is used to provide further understanding of the disclosure, constitutes part of this application, this public affairs
The illustrative embodiments and their description opened do not constitute the improper restriction to the disclosure for explaining the disclosure.In the accompanying drawings:
Fig. 1 is the flow diagram of the fall detection method according to the embodiment of the present disclosure 1;
Fig. 2 is the parameter of the used convolutional neural networks of the human body detecting method according to the embodiment of the present disclosure 1
Parameter list;
Fig. 3 is the schematic diagram that the human body detecting method according to the embodiment of the present disclosure 1 handles characteristic pattern;
Fig. 4 is the schematic diagram that the human body detecting method according to the embodiment of the present disclosure 1 handles characteristic pattern;
Fig. 5 is the parameter of the used convolutional neural networks of the fall detection method according to the embodiment of the present disclosure 1
Parameter list;
Fig. 6 is the schematic diagram of the device of the fall detection according to the embodiment of the present disclosure 2;
Fig. 7 is the schematic diagram of the device of the fall detection according to the embodiment of the present disclosure 3.
Specific embodiment
In order to make those skilled in the art more fully understand the technical solution of the disclosure, implement below in conjunction with the disclosure
Attached drawing in example, is clearly and completely described the technical solution in the embodiment of the present disclosure.Obviously, described embodiment
The only embodiment of disclosure a part, instead of all the embodiments.Based on the embodiment in the disclosure, this field is common
Disclosure protection all should belong in technical staff's every other embodiment obtained without making creative work
Range.
It should be noted that the specification and claims of the disclosure and term " first " in above-mentioned attached drawing, "
Two " etc. be to be used to distinguish similar objects, without being used to describe a particular order or precedence order.It should be understood that using in this way
Data be interchangeable under appropriate circumstances, so as to embodiment of the disclosure described herein can in addition to illustrating herein or
Sequence other than those of description is implemented.In addition, term " includes " and " having " and their any deformation, it is intended that cover
Cover it is non-exclusive include, for example, the process, method, system, product or equipment for containing a series of steps or units are not necessarily limited to
Step or unit those of is clearly listed, but may include be not clearly listed or for these process, methods, product
Or other step or units that equipment is intrinsic.
Embodiment 1
Fig. 1 shows the flow diagram of the fall detection method according to the present embodiment, and Fig. 2 shows the present embodiment
The parameter list of the parameter of the used convolutional neural networks of the human body detecting method;Fig. 3 is shown described in the present embodiment
Human body detecting method schematic diagram that characteristic pattern is handled;Fig. 4 shows human body detecting method pair described in the present embodiment
The schematic diagram that characteristic pattern is handled;Fig. 5 shows the used convolutional Neural of fall detection method described in the present embodiment
The parameter list of the parameter of network.
Refering to what is shown in Fig. 1, the fall detection method of the present embodiment the following steps are included:
S102: utilizing the human testing model based on convolutional neural networks, determines to include human body mesh in image to be detected
Target human body image region;And
S104: utilizing the fall detection model based on convolutional neural networks, according to human body image region, determines human body target
It is whether in a falling state.
Specifically, using the human testing model based on convolutional neural networks, determine to include human body in image to be detected
The human body image region of target.Then the fall detection model based on convolutional neural networks is utilized, according to determining human body image
Region determines whether the human body target in human body image region is in a falling state.
As described in the background art, the detection side of sensor is mainly based upon for tumble detection method for human body at present
The detection method of method, the detection method based on video image analysis and deep learning model analysis picture video.
Existing sensor-based detection method is based primarily upon wearable sensor, the movement of real-time measurement old man
Acceleration information or angular velocity information, then the information according to measurement judges whether old man falls.The shortcomings that such method is to need
Wearable sensor is carried, equipment cost is higher, and use is extremely inconvenient, and is easily lost, and area coverage is small, and market is general
And application cost is also higher.
The existing detection method based on video image analysis, by video background modeling technique, before obtaining moving target
Scape block is fallen by the prospect block feature of extraction, since there are many Moving Objects, such as various vehicles, various animals, Wu Fazhun
Really judge whether it is that people passes through, foreground target characteristic error is larger, causes erroneous detection and missing inspection more, and accuracy rate is low, is unable to reach
Commercial standard (CS).
The existing method based on deep learning model analysis picture video acquires scene image data by video camera,
Determine object detection area;According to scene image data, the instantaneous of all pixels point in current frame image is calculated with logic of propositions
Velocities field obtains pixel motion speed field picture;Assemble phase according to pixel motion speed field picture and default condition of similarity
Like pixel, Candidate Motion target area is formed, Candidate Motion target is obtained in Candidate Motion target area;Filter out candidate fortune
New moving target in moving-target;Pedestrian target is identified from new moving target according to deep learning method;Update pedestrian's mesh
Target target following list information;Variable condition and default Rule of judgment based on target following list information, judge pedestrian's mesh
Whether target falls and alarms.Method effect before this method is compared makes moderate progress, but algorithm flow is more complicated,
Consumption resource is larger, and operation time is longer, is difficult to be measured in real time tumble alarm, can not realize multi-path camera simultaneously
Detection operation, it is also difficult to be transplanted to mobile terminal and be detected, it is still higher to popularize hardware cost, is extremely difficult to commercial mark
It is quasi-.
Aiming at the problems existing in the prior art, the technical solution of the present embodiment provides a kind of human body fall detection side
Method uses the human testing model based on convolutional neural networks using processor, analyzes image, determines to include human body mesh
Target human body image region, and the fall detection model based on convolutional neural networks is used, according to the human body image area detected
Domain determines whether above-mentioned human body target is in a falling state.The technical solution of the application can be set from multiplex image acquisition simultaneously
Standby middle acquisition video data, obtains testing result using the detection model based on convolutional neural networks in real time.The inspection of the present embodiment
Survey a large amount of human body pictures of neural network learning visual signature and tumble human body picture posture feature, including hand, elbow joint,
The visual signature at the positions such as shoulder, back, waist, stern, knee joint, foot, and relationship when tumble between each genius loci, accuracy rate
It is high.And since this programme uses the detection model based on convolutional neural networks, not relative to traditional detection method
Sensor is needed, easy to use, calculating occupancy resource is not high, and cost is relatively low, is easy to universal.
To which the technical solution of the present embodiment solves detection accuracy present in existing tumble detection method for human body not
High, inconvenient for use, higher cost calculates the technical problem for occupying higher, the universal difficulty of resource.
In addition, for example, the technical solution of the present embodiment when obtaining image to be detected, such as the figure such as can use camera
As acquisition equipment, and image to be detected is sent to processor.Wherein image to be detected for example can be derived from camera acquisition
Video, processor obtain video data from video flowing, and are image frame data by video data decoding, wherein image frame data
As above-mentioned image to be detected.
Optionally, using the human testing model based on convolutional neural networks, determine to include human body from image to be detected
The operation in the human body image region of target, comprising: utilize human testing model, according to image to be detected, generate respectively with it is to be checked
The corresponding multiple vectors of multiple rectangle frame regions in altimetric image, wherein vector includes at least following information: corresponding rectangle frame
It include the confidence of human body target in the location information in region, the dimension information of corresponding rectangle frame region and corresponding rectangle frame
Spend information;And rectangle frame region corresponding to the maximum vector of confidence information is determined as human body image region.
Specifically, using the human testing model based on convolutional neural networks, according to image to be detected, generate respectively with to
The corresponding multiple vectors of multiple rectangle frame regions in detection image.Concrete example as illustrated with reference to fig. 2, in the network structure of Fig. 2
In, input is the image of 448*448, first passes through the characteristic pattern that convolutional layer generates 14*14*256.Then it is up-sampled twice,
The characteristic pattern for ultimately generating " 56*56*15 " exports the characteristic pattern of 15 56*56.Wherein in " 56*56*15 ", " 56*56 " table
After showing that original image reduces 8 times, the characteristic pattern of 56*56 is generated, the characteristic pattern of generation is used to predict the target of different scale.Wherein
Different scale for example can be large, medium and small three scales.
In addition, the meaning of last one-dimensional " 15 " is (x, y, w, h, Confidence) * in the characteristic pattern of " 56*56*15 "
3, i.e., 15 characteristic patterns are divided into 3 groups, every group include 5 56*56 characteristic pattern.Refering to what is shown in Fig. 3, in each group of characteristic pattern
In, the element (such as black dot at the same position marked in figure) of same position respectively indicates in same 5 dimensional vector
An element.In this way, each group of characteristic pattern means that 56*56 5 dimensional vectors.To which 15 characteristic patterns have meant that 3*56*
56 5 dimensional vectors.
Wherein, " x " indicates the rectangle frame region left margin in image to be detected in the i-th row, jth column grid (i, j)
Horizontal offset values divided by mesh width multiple;" y " indicate image to be detected in rectangle frame region coboundary the i-th row,
In jth column grid (i, j) vertical offset value divided by grid height multiple;" w " indicates the width in image to be detected divided by net
The multiple of lattice width;" h " indicates the height in image to be detected divided by the multiple of network height;" Confidence " indicates to be checked
The confidence score of rectangle frame region in altimetric image.
To, using human testing model, according to image to be detected, generate respectively with the rectangle frame area in image to be detected
The corresponding vector in domain is (x, y, w, h, Confidence).Wherein, x, y indicate the location information of corresponding rectangle frame region, w, h table
Show that the dimension information of corresponding rectangle frame region, Confidence indicate that the confidence level in corresponding rectangle frame comprising human body target is believed
Breath.
It further, can be by square corresponding to the vector of confidence information highest scoring in above-mentioned image to be detected
Shape frame region is determined as human body image region.To may further determine that the people in image after determining human body image region
Body coordinate (x, y, w, h).Wherein, " x " indicates the rectangle frame region left margin in image to be detected in the i-th row, jth column grid
Horizontal offset values in (i, j) divided by mesh width multiple;" y " indicates the rectangle frame region in image to be detected in coboundary
In the i-th row, jth column grid (i, j) vertical offset value divided by grid height multiple;" w " indicates the width in image to be detected
Spend the multiple divided by mesh width;" h " indicates the height in image to be detected divided by the multiple of network height.
Optionally, the operation of multiple vectors corresponding with multiple rectangle frame regions in image to be detected respectively, packet are generated
It includes: multiple eigenmatrixes being generated according to image to be detected using human testing model;And using in multiple eigenmatrixes
The element of the same position of at least part eigenmatrix constructs a vector in multiple vectors.
Specifically, multiple eigenmatrixes are generated according to image to be detected using human testing model.Specifically for example in Fig. 2
Network structure in, input is the image of 448*448, and up-sampling ultimately generates " 56*56*15 " by convolutional layer and twice
Characteristic pattern.Wherein in " 56*56*15 ", after " 56*56 " indicates that original image reduces 8 times, the characteristic pattern of 56*56 is generated, i.e., most
Throughout one's life at the characteristic pattern (56*56*15) of 15 56*56.15 characteristic patterns are divided into 3 groups, every group include 5 56*56 feature
Figure.Refering to what is shown in Fig. 3, in each group of characteristic pattern, the element of same position (such as the black at the same position marked in figure
Dot) respectively indicate an element in same 5 dimensional vector.In this way, each group of characteristic pattern means that 56*56 5 dimensional vectors.
To which 15 characteristic patterns have meant that 3*56*56 5 dimensional vectors.
Optionally, the element of the same position of at least part eigenmatrix in multiple eigenmatrixes constructs described more
The operation of a vector in a vector, comprising: multiple eigenmatrixes are divided into multiple set of matrices, wherein each matrix stack
The quantity for the eigenmatrix for including in conjunction is identical;And the member of the same position using the eigenmatrix in the same set of matrices
Element constructs a vector in multiple vectors.
Specifically, refering to what is shown in Fig. 4,15 characteristic patterns are divided into 3 groups, every group include 5 56*56 characteristic pattern.Every
In one group of characteristic pattern, the element (such as black dot at the same position marked in figure) of same position respectively indicates same
An element in 5 dimensional vectors.In this way, each group of characteristic pattern means that 56*56 5 dimensional vectors.Every group of characteristic pattern prediction is different
The target of scale.Refering to what is shown in Fig. 4, wherein different scale for example can be large, medium and small three scales.Wherein, large, medium and small three
A scale for example can be the prefabricated template of 28*28,56*56,112*112.
For example, the corresponding rectangle frame region in human body image region carries out returning its correspondence according to the prefabricated template of 28*28
Coordinate vector when, such as true human body rectangle frame be 25*32, then the corresponding rectangle frame region of the human body target returned out
The w value of vector be 0.8925557, h value be 1.142857, so as to return out the corresponding rectangle frame region of human body target
The corresponding vector of coordinate.
It is thus possible to complete the detection to the human body target in image to be detected.
Optionally, human body mesh is determined according to human body image region using the fall detection model based on convolutional neural networks
Mark whether operation in a falling state, comprising: utilize the fall detection model based on convolutional neural networks, according to human body image
Region generates the fractional value whether fallen for identifying human body target;And when fractional value is greater than predetermined threshold, determine human body
It is in a falling state.
Specifically, it is generated according to human body image region for marking using the fall detection model based on convolutional neural networks
It further include that image size is changed by fall detection model, wherein changing figure before knowing the fractional value whether human body target falls
The input feature vector figure of fall detection model is transformed to by the present embodiment by image interpolation as the mode of size for example can be
Fall detection model inputs size.
Refering to what is shown in Fig. 5, the input picture of fall detection neural network passes through image interpolation in the network structure of Fig. 5
After changing size, the input picture of 112*112 is generated.Then after the compression for passing through convolutional layer, full articulamentum and average pond,
Finally by softmax layers of prediction output valve.This output valve is as used to identify the fractional value whether human body target falls, wherein
The value range of fractional value is 0~1.When doing tumble state and judging, such as a predetermined threshold can be set, wherein predetermined threshold
The value of value for example can be 0.5, and when the fractional value of above-mentioned output is greater than 0.5, i.e. judgement human body is in a falling state, otherwise
Determine that human body is not on tumble state.
Optionally, it is generated according to human body image region for marking using the fall detection model based on convolutional neural networks
Know the operation for the fractional value whether human body target falls, comprising: using the neural network structure in fall detection model, extract people
Characteristics of image in body image-region;And using the taxonomic structure in fall detection model, according to extracted characteristics of image,
Generate the fractional value whether fallen for identifying human body target.
Specifically, refering to what is shown in Fig. 5, using fall detection model, by the convolutional layer of neural network structure, human body is extracted
It is raw according to extracted characteristics of image using the taxonomic structure in fall detection model after characteristics of image in image-region
At the fractional value whether fallen for identifying human body target.
Optionally, it is generated according to human body image region for marking using the fall detection model based on convolutional neural networks
The operation for knowing the fractional value whether human body target falls, further includes the fractional value for converting fractional value to Probability Forms.
Specifically, convolutional layer is passed through using the fall detection model based on convolutional neural networks according to human body image region
After extracting the characteristics of image in human body image region, after full articulamentum, it is converted by softmax layers of contribute's fractional value
The fractional value of Probability Forms, value range are 0~1.So as to determine whether human body target is according to above-mentioned fractional value
Tumble state.
In addition, construct the present embodiment scheme in human testing model when, acquire first sufficient amount of the first
Body image data, wherein the first human body image data includes with different country origins, the different colour of skin, different sexes, all ages and classes, no
The human body of same posture, different scenes;Then the first human body image data is cleaned, and marked in the first human body image data
Human body minimum circumscribed rectangle frame coordinate (x, y, w, h);It finally arranges the training set of the first human body image data, verifying collection, survey
Examination collection.
Wherein, the human testing model in the scheme of the present embodiment is constructed further include: acquire the first new human figure first
Sheet data, wherein the first new human body image data includes with different country origins, the different colour of skin, different sexes, all ages and classes, no
The human body of same posture, different scenes;Then the first new human body image data is detected using the model of above-mentioned training set,
And the first new human body image data that will test result inaccuracy is marked again;Finally arrange the first new human body picture
Training set, the verifying collection, test set of data, to obtain the human testing model of the present embodiment.
In addition, construct the present embodiment scheme in fall detection model when, acquire sufficient amount of second people first
Body image data;Wherein the second human body image data includes the human body with tumble posture;It marks in the second human body image data
Human body minimum circumscribed rectangle frame coordinate (x, y, w, h), and cut out the picture in human body minimum circumscribed rectangle frame;To cutting out
Human body minimum circumscribed rectangle frame in picture carry out classification annotation, such as 0 be normal picture, 1 be tumble picture;Finally arrange
Training set, the verifying collection, test set of second human body image data.
Wherein, the fall detection model in the scheme of the present embodiment is constructed further include: acquire second new human figure's the piece number
According to wherein the first new human body image data includes having tumble posture and the human body with non-tumble posture;Utilize above-mentioned instruction
The model for practicing collection carries out classification annotation, and the second new human body that classification results are inaccurate to the second new human body image data
Image data is marked again;Training set, the verifying collection, test set of the second new human body image data are finally arranged, thus
Obtain the fall detection model of the present embodiment.
To use the human testing based on convolutional neural networks using processor according to the technical solution of the present embodiment
Model analyzes image, determines the human body image region comprising human body target, and use falling based on convolutional neural networks
Detection model determines whether above-mentioned human body target is in a falling state according to the human body image region detected.The application's
Technical solution can obtain video data simultaneously from multiplex image acquisition equipment, and can be used based on convolutional neural networks
Detection model show that testing result, accuracy rate are high in real time.And since this programme uses the detection based on convolutional neural networks
Model, therefore sensor is not needed relative to traditional detection method, easy to use, calculating occupancy resource is not high, and cost is relatively low,
It is easy to universal.
In addition, in order to make it easy to understand, as follows to the supplementary explanation of the technical solution of the present embodiment step in chronological order.
Present embodiments provide following technical solution:
Step 1: the human testing model in the scheme of the present embodiment is constructed.Sufficient amount of first human body is acquired first
Image data, wherein the first human body image data includes with different country origins, the different colours of skin, different sexes, all ages and classes, difference
The human body of posture, different scenes;Then the first human body image data is cleaned, and marked in the first human body image data
Human body minimum circumscribed rectangle frame coordinate (x, y, w, h);Finally arrange training set, the verifying collection, test of the first human body image data
Collection.
Step 2: the fall detection model in the scheme of the present embodiment is constructed.Sufficient amount of second human body is acquired first
Image data;Wherein the second human body image data includes the human body with tumble posture;It marks in the second human body image data
Human body minimum circumscribed rectangle frame coordinate (x, y, w, h), and cut out the picture in human body minimum circumscribed rectangle frame;To what is cut out
Picture in human body minimum circumscribed rectangle frame carries out classification annotation;Finally arrange training set, the verifying of the second human body image data
Collection, test set.
Step 3: building human testing neural network, returns out human body coordinate (x, y, w, h), utilizes the first moment of gradient
Battle array estimation and second-order matrix estimate the accuracy of combination training pattern, the parameter of used neural network structure such as Fig. 2 institute
Show.Input is the image of 448*448, up-samples by convolutional layer and twice the characteristic pattern for ultimately generating " 56*56*15 ".Its
In in " 56*56*15 ", " 56*56 " indicate original image reduce 8 times after, generate the characteristic pattern of 56*56, the characteristic pattern of generation
For predicting the target of different scale.Wherein different scale for example can be large, medium and small three scales.
The meaning of last one-dimensional " 15 " is (x, y, w, h, Confidence) * 3, i.e., 15 characteristic patterns is divided into 3 groups, every group
Characteristic pattern including 5 56*56.Wherein, " x " indicates the rectangle frame region left margin in image to be detected in the i-th row, the
Horizontal offset values in j column grid (i, j) divided by mesh width multiple;" y " indicates the rectangle frame region in image to be detected
Coboundary in the i-th row, jth column grid (i, j) vertical offset value divided by grid height multiple;" w " indicates mapping to be checked
As in width divided by mesh width multiple;" h " indicates the height in image to be detected divided by the multiple of network height;
" Confidence " indicates the confidence score of the rectangle frame region in image to be detected.
To choose the rectangle frame region of objective degrees of confidence highest scoring according to the value of Confidence, can further count
Calculate the coordinate (x, y, w, h) of the corresponding vector of rectangle frame region of human body target.
Step 4: building fall detection neural network, to the to be detected with human body target of human testing network output
Picture carries out tumble state-detection, that is, detects whether to be estimated using the first order matrix estimation of gradient and second-order matrix for tumble state
The accuracy of combination training pattern is counted, the parameter of used neural network structure is as shown in Figure 5.Fall detection neural network
Input picture after image interpolation changes size, after the compression by convolutional layer, full articulamentum and average pond, finally
Pass through softmax layers of prediction output valve.This output valve is to be used to identify the fractional value whether human body target falls, mid-score
The value range of value is 0~1.So as to determine whether human body target is in a falling state according to above-mentioned fractional value.
Step 5: acquiring the first new human body image data, wherein the first new human body image data includes having difference
Country origin, the different colour of skin, different sexes, all ages and classes, different gestures, different scenes human body;Then it is trained using step 3
Human testing model the first new human body image data is detected, and will test the first new human body of result inaccuracy
Image data is marked again;Training set, the verifying collection, test set for finally arranging the first new human body image data, until
Until accuracy rate reaches requirement.
Step 6: step 5 is repeated, until million picture accuracys rate reach requirement deconditioning.
Step 7: acquiring the second new human body image data, wherein the first new human body image data includes having to fall
Posture and human body with non-tumble posture;Using the trained fall detection model of step 4 to second new human figure's the piece number
It is marked again according to progress classification annotation, and by the second new human body image data of classification results inaccuracy;Finally arrange
Training set, the verifying collection, test set of the second new human body image data, until accuracy rate reaches requirement.
Step 8: step 7 is repeated, until million picture test accuracy rates reach requirement deconditioning.
Step 9: obtaining data from video flowing, is decoded into image frame data with decoder module
Step 10: with the human body minimum circumscribed rectangle frame in the human testing model prediction image frame data of pre-training, from
And complete detection human body target.
Step 11: there is the human body target in the image data of human body target with the fall detection model inspection of pre-training
It whether is tumble state, and output test result, wherein testing result can for example be shown by being sent to terminal.
In addition, providing a kind of storage medium according to the second aspect of the present embodiment, the storage medium includes storage
Program, wherein described program operation when as processor execute any of the above one described in method.
Through the above description of the embodiments, those skilled in the art can be understood that according to above-mentioned implementation
The method of example can be realized by means of software and necessary general hardware platform, naturally it is also possible to by hardware, but it is very much
In the case of the former be more preferably embodiment.Based on this understanding, technical solution of the present invention is substantially in other words to existing
The part that technology contributes can be embodied in the form of software products, which is stored in a storage
In medium (such as ROM/RAM, magnetic disk, CD), including some instructions are used so that a terminal device (can be mobile phone, calculate
Machine, server or network equipment etc.) execute method described in each embodiment of the present invention.
Embodiment 2
Fig. 6 shows the schematic diagram of the device of fall detection described in the present embodiment.The falling detection device of the present embodiment
600 is corresponding with the method according to embodiment 1.
Refering to what is shown in Fig. 6, the device 600 includes: human body image area determination module 610, for using based on convolution mind
Human testing model through network determines the human body image region including human body target in image to be detected;And tumble shape
State determination module 620, according to human body image region, determines people for utilizing the fall detection model based on convolutional neural networks
Whether body target is in a falling state.
Optionally, human body image area determination module 610 includes: that vector generates submodule, for using human testing model,
According to image to be detected, multiple vectors corresponding with multiple rectangle frame regions in image to be detected respectively are generated, wherein vector
Including at least following information: the location information of corresponding rectangle frame region, the dimension information of corresponding rectangle frame region and right
It include the confidence information of human body target in the rectangle frame answered;And human body image region determines submodule, is used for confidence level
Rectangle frame region corresponding to the maximum vector of information is determined as human body image region.
Optionally, it includes: eigenmatrix generation unit that vector, which generates submodule, for utilizing human testing model, according to
Image to be detected generates multiple eigenmatrixes;And vector construction unit, for utilizing at least one in multiple eigenmatrixes
The element for dividing the same position of eigenmatrix, constructs a vector in multiple vectors.
Optionally, vector construction unit includes: that set of matrices divides subelement, more for being divided into multiple eigenmatrixes
A set of matrices, wherein the quantity for the eigenmatrix for including in each set of matrices is identical;And vector constructs subelement, is used for
Using the element of the same position of the eigenmatrix in the same set of matrices, a vector in multiple vectors is constructed.
Optionally, tumble state determination module 620 includes: that fractional value generates submodule, for using based on convolutional Neural
The fall detection model of network generates the fractional value whether fallen for identifying human body target according to human body image region;And
Tumble state decision sub-module, for determining that human body is in a falling state when fractional value is greater than predetermined threshold.
Optionally, it includes: image characteristics extraction unit that fractional value, which generates submodule, for using in fall detection model
Neural network structure extracts the characteristics of image in human body image region;And fractional value generation unit, for utilizing fall detection
Taxonomic structure in model generates the fractional value whether fallen for identifying human body target according to extracted characteristics of image.
Optionally, fractional value generation unit includes fractional value transforming subunit, for converting Probability Forms for fractional value
Fractional value.
To use the human testing based on convolutional neural networks using processor according to the technical solution of the present embodiment
Model analyzes image, determines the human body image region comprising human body target, and use falling based on convolutional neural networks
Detection model determines whether above-mentioned human body target is in a falling state according to the human body image region detected.The application's
Technical solution can obtain video data simultaneously from multiplex image acquisition equipment, and can be used based on convolutional neural networks
Detection model show that testing result, accuracy rate are high in real time.And since this programme uses the detection based on convolutional neural networks
Model, therefore sensor is not needed relative to traditional detection method, easy to use, calculating occupancy resource is not high, and cost is relatively low,
It is easy to universal.
Therefore, the technical solution of the present embodiment solves detection accuracy present in existing tumble detection method for human body not
High, inconvenient for use, higher cost calculates the technical problem for occupying higher, the universal difficulty of resource.
Embodiment 3
Fig. 7 shows the schematic diagram of the device of fall detection described in the present embodiment.The falling detection device of the present embodiment
700 is corresponding with the method according to embodiment 1.
Refering to what is shown in Fig. 7, the device 700 includes: processor 710;And memory 720, it connect, uses with processor 710
In providing the instruction for handling following processing step for processor 710: utilizing the human testing model based on convolutional neural networks, In
The human body image region including human body target is determined in image to be detected;And utilize the fall detection based on convolutional neural networks
Model determines whether human body target is in a falling state according to human body image region.
Optionally, memory 720 is also used to provide the instruction for handling following processing step for processor 710: utilizing human body
Detection model generates multiple vectors corresponding with multiple rectangle frame regions in image to be detected respectively according to image to be detected,
Wherein vector includes at least following information: the size letter of the location information of corresponding rectangle frame region, corresponding rectangle frame region
It include the confidence information of human body target in breath and corresponding rectangle frame;And it will be corresponding to the maximum vector of confidence information
Rectangle frame region be determined as human body image region.
Optionally, memory 720 is also used to provide the instruction for handling following processing step for processor 710: utilizing human body
Detection model generates multiple eigenmatrixes according to image to be detected;And it is special using at least part in multiple eigenmatrixes
The element for levying the same position of matrix, constructs a vector in multiple vectors.
Optionally, memory 720 is also used to provide the instruction for handling following processing step for processor 710: by multiple spies
Sign matrix is divided into multiple set of matrices, wherein the quantity for the eigenmatrix for including in each set of matrices is identical;And it utilizes
The element of the same position of eigenmatrix in the same set of matrices constructs a vector in multiple vectors.
Optionally, memory 720 is also used to provide the instruction for handling following processing step for processor 710: using being based on
The fall detection model of convolutional neural networks generates point whether fallen for identifying human body target according to human body image region
Numerical value;And when fractional value is greater than predetermined threshold, determine that human body is in a falling state.
Optionally, memory 720 is also used to provide the instruction for handling following processing step for processor 710: utilizing tumble
Neural network structure in detection model extracts the characteristics of image in human body image region;And using in fall detection model
Taxonomic structure the fractional value whether fallen for identifying the human body target is generated according to extracted characteristics of image.
Optionally, memory 720 is also used to provide the instruction for handling following processing step for processor 710: by fractional value
It is converted into the fractional value of Probability Forms.
To use the human testing based on convolutional neural networks using processor according to the technical solution of the present embodiment
Model analyzes image, determines the human body image region comprising human body target, and use falling based on convolutional neural networks
Detection model determines whether above-mentioned human body target is in a falling state according to the human body image region detected.The application's
Technical solution can obtain video data simultaneously from multiplex image acquisition equipment, and can be used based on convolutional neural networks
Detection model show that testing result, accuracy rate are high in real time.And since this programme uses the detection based on convolutional neural networks
Model, therefore sensor is not needed relative to traditional detection method, easy to use, calculating occupancy resource is not high, and cost is relatively low,
It is easy to universal.
Therefore, the technical solution of the present embodiment solves detection accuracy present in existing tumble detection method for human body not
High, inconvenient for use, higher cost calculates the technical problem for occupying higher, the universal difficulty of resource.
The serial number of the above embodiments of the invention is only for description, does not represent the advantages or disadvantages of the embodiments.
In the above embodiment of the invention, it all emphasizes particularly on different fields to the description of each embodiment, does not have in some embodiment
The part of detailed description, reference can be made to the related descriptions of other embodiments.
In several embodiments provided herein, it should be understood that disclosed technology contents can pass through others
Mode is realized.Wherein, the apparatus embodiments described above are merely exemplary, such as the division of the unit, only
A kind of logical function partition, there may be another division manner in actual implementation, for example, multiple units or components can combine or
Person is desirably integrated into another system, or some features can be ignored or not executed.Another point, shown or discussed is mutual
Between coupling, direct-coupling or communication connection can be through some interfaces, the INDIRECT COUPLING or communication link of unit or module
It connects, can be electrical or other forms.
The unit as illustrated by the separation member may or may not be physically separated, aobvious as unit
The component shown may or may not be physical unit, it can and it is in one place, or may be distributed over multiple
In network unit.It can select some or all of unit therein according to the actual needs to realize the mesh of this embodiment scheme
's.
It, can also be in addition, the functional units in various embodiments of the present invention may be integrated into one processing unit
It is that each unit physically exists alone, can also be integrated in one unit with two or more units.Above-mentioned integrated list
Member both can take the form of hardware realization, can also realize in the form of software functional units.
If the integrated unit is realized in the form of SFU software functional unit and sells or use as independent product
When, it can store in a computer readable storage medium.Based on this understanding, technical solution of the present invention is substantially
The all or part of the part that contributes to existing technology or the technical solution can be in the form of software products in other words
It embodies, which is stored in a storage medium, including some instructions are used so that a computer
Equipment (can for personal computer, server or network equipment etc.) execute each embodiment the method for the present invention whole or
Part steps.And storage medium above-mentioned includes: that USB flash disk, read-only memory (ROM, Read-Only Memory), arbitrary access are deposited
Reservoir (RAM, Random Access Memory), mobile hard disk, magnetic or disk etc. be various to can store program code
Medium.
The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also answered
It is considered as protection scope of the present invention.
Claims (10)
1. a kind of fall detection method, which comprises the following steps:
Using the human testing model based on convolutional neural networks, the human figure including human body target is determined in image to be detected
As region;And
The human body target is determined according to the human body image region using the fall detection model based on convolutional neural networks
It is whether in a falling state.
2. fall detection method according to claim 1, which is characterized in that examined using the human body based on convolutional neural networks
Model is surveyed, the operation in the human body image region including human body target is determined from image to be detected, comprising:
Using the human testing model, according to described image to be detected, generate respectively with it is multiple in described image to be detected
The corresponding multiple vectors of rectangle frame region, wherein the vector includes at least following information: the position of corresponding rectangle frame region
It include the confidence information of human body target in information, the dimension information of corresponding rectangle frame region and corresponding rectangle frame;With
And
Rectangle frame region corresponding to the maximum vector of the confidence information is determined as the human body image region.
3. fall detection method according to claim 2, which is characterized in that generate respectively and in described image to be detected
The operation of the corresponding multiple vectors of multiple rectangle frame regions, comprising:
Multiple eigenmatrixes are generated according to described image to be detected using the human testing model;And
Using the element of the same position of at least part eigenmatrix in the multiple eigenmatrix, construct it is the multiple to
A vector in amount.
4. fall detection method according to claim 3, which is characterized in that utilize in the multiple eigenmatrix at least
The element of the same position of a part of eigenmatrix constructs the operation of a vector in the multiple vector, comprising:
The multiple eigenmatrix is divided into multiple set of matrices, wherein the number for the eigenmatrix for including in each set of matrices
It measures identical;And
Using the element of the same position of the eigenmatrix in the same set of matrices, construct one in the multiple vector to
Amount.
5. fall detection method according to claim 1, which is characterized in that examined using the tumble based on convolutional neural networks
It surveys model and human body target operation whether in a falling state is determined according to the human body image region, comprising:
Using the fall detection model based on convolutional neural networks, according to the human body image region, generate described for identifying
The fractional value whether human body target falls;And
When the fractional value is greater than predetermined threshold, determine that the human body is in a falling state.
6. fall detection method according to claim 5, which is characterized in that examined using the tumble based on convolutional neural networks
It surveys model and the operation for identifying the fractional value whether human body target falls is generated according to the human body image region, wrap
It includes:
Using the neural network structure in the fall detection model, the characteristics of image in the human body image region is extracted;With
And
It is generated according to extracted characteristics of image for identifying the people using the taxonomic structure in the fall detection model
The fractional value whether body target falls.
7. fall detection method according to claim 6, which is characterized in that examined using the tumble based on convolutional neural networks
It surveys model and the operation for identifying the fractional value whether human body target falls is generated according to the human body image region, also
Fractional value including converting the fractional value to Probability Forms.
8. a kind of storage medium, which is characterized in that the storage medium includes the program of storage, wherein run in described program
When as processor perform claim require any one of 1 to 7 described in method.
9. a kind of falling detection device characterized by comprising
Human body image area determination module, for utilizing the human testing model based on convolutional neural networks, in image to be detected
Middle determination includes the human body image region of human body target;And
Tumble state determination module, for utilizing the fall detection model based on convolutional neural networks, according to the human body image
Region determines whether the human body target is in a falling state.
10. a kind of falling detection device characterized by comprising
Processor;And
Memory is connected to the processor, for providing the instruction for handling following processing step for the processor:
Using the human testing model based on convolutional neural networks, the human figure including human body target is determined in image to be detected
As region;And
The human body target is determined according to the human body image region using the fall detection model based on convolutional neural networks
It is whether in a falling state.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910621095.1A CN110443150A (en) | 2019-07-10 | 2019-07-10 | A kind of fall detection method, device, storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910621095.1A CN110443150A (en) | 2019-07-10 | 2019-07-10 | A kind of fall detection method, device, storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110443150A true CN110443150A (en) | 2019-11-12 |
Family
ID=68430120
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910621095.1A Pending CN110443150A (en) | 2019-07-10 | 2019-07-10 | A kind of fall detection method, device, storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110443150A (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111241913A (en) * | 2019-12-19 | 2020-06-05 | 北京文安智能技术股份有限公司 | Method, device and system for detecting falling of personnel |
CN111461042A (en) * | 2020-04-07 | 2020-07-28 | 中国建设银行股份有限公司 | Fall detection method and system |
CN111507252A (en) * | 2020-04-16 | 2020-08-07 | 上海眼控科技股份有限公司 | Human body falling detection device and method, electronic terminal and storage medium |
WO2021104007A1 (en) * | 2019-11-26 | 2021-06-03 | 京东数科海益信息科技有限公司 | Method and device for animal state monitoring, electronic device, and storage medium |
CN113221661A (en) * | 2021-04-14 | 2021-08-06 | 浪潮天元通信信息系统有限公司 | Intelligent human body tumbling detection system and method |
CN113642361A (en) * | 2020-05-11 | 2021-11-12 | 杭州萤石软件有限公司 | Method and equipment for detecting falling behavior |
CN117636404A (en) * | 2024-01-26 | 2024-03-01 | 贵州信邦富顿科技有限公司 | Fall detection method and system based on non-wearable equipment |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103211599A (en) * | 2013-05-13 | 2013-07-24 | 桂林电子科技大学 | Method and device for monitoring tumble |
CN105279483A (en) * | 2015-09-28 | 2016-01-27 | 华中科技大学 | Fall-down behavior real-time detection method based on depth image |
CN107220604A (en) * | 2017-05-18 | 2017-09-29 | 清华大学深圳研究生院 | A kind of fall detection method based on video |
CN108062531A (en) * | 2017-12-25 | 2018-05-22 | 南京信息工程大学 | A kind of video object detection method that convolutional neural networks are returned based on cascade |
CN108090458A (en) * | 2017-12-29 | 2018-05-29 | 南京阿凡达机器人科技有限公司 | Tumble detection method for human body and device |
CN108154113A (en) * | 2017-12-22 | 2018-06-12 | 重庆邮电大学 | Tumble event detecting method based on full convolutional network temperature figure |
CN108805093A (en) * | 2018-06-19 | 2018-11-13 | 华南理工大学 | Escalator passenger based on deep learning falls down detection algorithm |
CN109299703A (en) * | 2018-10-17 | 2019-02-01 | 思百达物联网科技(北京)有限公司 | The method, apparatus and image capture device counted to mouse feelings |
CN109800860A (en) * | 2018-12-28 | 2019-05-24 | 北京工业大学 | A kind of Falls in Old People detection method of the Community-oriented based on CNN algorithm |
-
2019
- 2019-07-10 CN CN201910621095.1A patent/CN110443150A/en active Pending
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103211599A (en) * | 2013-05-13 | 2013-07-24 | 桂林电子科技大学 | Method and device for monitoring tumble |
CN105279483A (en) * | 2015-09-28 | 2016-01-27 | 华中科技大学 | Fall-down behavior real-time detection method based on depth image |
CN107220604A (en) * | 2017-05-18 | 2017-09-29 | 清华大学深圳研究生院 | A kind of fall detection method based on video |
CN108154113A (en) * | 2017-12-22 | 2018-06-12 | 重庆邮电大学 | Tumble event detecting method based on full convolutional network temperature figure |
CN108062531A (en) * | 2017-12-25 | 2018-05-22 | 南京信息工程大学 | A kind of video object detection method that convolutional neural networks are returned based on cascade |
CN108090458A (en) * | 2017-12-29 | 2018-05-29 | 南京阿凡达机器人科技有限公司 | Tumble detection method for human body and device |
CN108805093A (en) * | 2018-06-19 | 2018-11-13 | 华南理工大学 | Escalator passenger based on deep learning falls down detection algorithm |
CN109299703A (en) * | 2018-10-17 | 2019-02-01 | 思百达物联网科技(北京)有限公司 | The method, apparatus and image capture device counted to mouse feelings |
CN109800860A (en) * | 2018-12-28 | 2019-05-24 | 北京工业大学 | A kind of Falls in Old People detection method of the Community-oriented based on CNN algorithm |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2021104007A1 (en) * | 2019-11-26 | 2021-06-03 | 京东数科海益信息科技有限公司 | Method and device for animal state monitoring, electronic device, and storage medium |
CN111241913A (en) * | 2019-12-19 | 2020-06-05 | 北京文安智能技术股份有限公司 | Method, device and system for detecting falling of personnel |
CN111461042A (en) * | 2020-04-07 | 2020-07-28 | 中国建设银行股份有限公司 | Fall detection method and system |
CN111507252A (en) * | 2020-04-16 | 2020-08-07 | 上海眼控科技股份有限公司 | Human body falling detection device and method, electronic terminal and storage medium |
CN113642361A (en) * | 2020-05-11 | 2021-11-12 | 杭州萤石软件有限公司 | Method and equipment for detecting falling behavior |
WO2021227874A1 (en) * | 2020-05-11 | 2021-11-18 | 杭州萤石软件有限公司 | Falling behaviour detection method and device |
CN113642361B (en) * | 2020-05-11 | 2024-01-23 | 杭州萤石软件有限公司 | Fall behavior detection method and equipment |
CN113221661A (en) * | 2021-04-14 | 2021-08-06 | 浪潮天元通信信息系统有限公司 | Intelligent human body tumbling detection system and method |
CN117636404A (en) * | 2024-01-26 | 2024-03-01 | 贵州信邦富顿科技有限公司 | Fall detection method and system based on non-wearable equipment |
CN117636404B (en) * | 2024-01-26 | 2024-04-16 | 贵州信邦富顿科技有限公司 | Fall detection method and system based on non-wearable equipment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110443150A (en) | A kind of fall detection method, device, storage medium | |
CN111881705B (en) | Data processing, training and identifying method, device and storage medium | |
JP6522060B2 (en) | Object recognition device, classification tree learning device and operation method thereof | |
CN105740780A (en) | Method and device for human face in-vivo detection | |
CN105022982B (en) | Hand motion recognition method and apparatus | |
CN110175993A (en) | A kind of Faster R-CNN pulmonary tuberculosis sign detection system and method based on FPN | |
CN107808143A (en) | Dynamic gesture identification method based on computer vision | |
CN109635875A (en) | A kind of end-to-end network interface detection method based on deep learning | |
CN109191588A (en) | Move teaching method, device, storage medium and electronic equipment | |
CN110532874B (en) | Object attribute recognition model generation method, storage medium and electronic device | |
CN109670380A (en) | Action recognition, the method and device of pose estimation | |
CN115661943B (en) | Fall detection method based on lightweight attitude assessment network | |
CN113111767A (en) | Fall detection method based on deep learning 3D posture assessment | |
WO2017161734A1 (en) | Correction of human body movements via television and motion-sensing accessory and system | |
CN112861723B (en) | Sports action recognition counting method and device based on human body gesture recognition and computer readable storage medium | |
CN110298279A (en) | A kind of limb rehabilitation training householder method and system, medium, equipment | |
CN105912991A (en) | Behavior identification method based on 3D point cloud and key bone nodes | |
CN110084192A (en) | Quick dynamic hand gesture recognition system and method based on target detection | |
KR20180110443A (en) | Apparatus and method for providing calorie information | |
CN110321820A (en) | A kind of sight drop point detection method based on contactless device | |
US20220415091A1 (en) | Ml model arrangement and method for evaluating motion patterns | |
CN111883229A (en) | Intelligent movement guidance method and system based on visual AI | |
CN115294660B (en) | Body-building action recognition model, training method of model and body-building action recognition method | |
Ferreira et al. | Deep learning approaches for workout repetition counting and validation | |
CN103455826B (en) | Efficient matching kernel body detection method based on rapid robustness characteristics |
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: 20191112 |