CN106446876B - Sensing behavior identification method and device - Google Patents
Sensing behavior identification method and device Download PDFInfo
- Publication number
- CN106446876B CN106446876B CN201611024234.5A CN201611024234A CN106446876B CN 106446876 B CN106446876 B CN 106446876B CN 201611024234 A CN201611024234 A CN 201611024234A CN 106446876 B CN106446876 B CN 106446876B
- Authority
- CN
- China
- Prior art keywords
- target information
- sensing
- information matrix
- layer
- data
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 32
- 239000011159 matrix material Substances 0.000 claims abstract description 116
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 38
- 239000011551 heat transfer agent Substances 0.000 claims description 31
- 230000000694 effects Effects 0.000 claims description 23
- 238000004458 analytical method Methods 0.000 claims description 18
- 238000012937 correction Methods 0.000 claims description 12
- 230000003542 behavioural effect Effects 0.000 claims description 10
- 238000013528 artificial neural network Methods 0.000 claims description 8
- 229910002056 binary alloy Inorganic materials 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 238000003491 array Methods 0.000 abstract description 2
- 230000006399 behavior Effects 0.000 description 30
- 238000010586 diagram Methods 0.000 description 10
- 238000012549 training Methods 0.000 description 10
- 230000004044 response Effects 0.000 description 5
- 230000008859 change Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000005286 illumination Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000037081 physical activity Effects 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000006698 induction Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000001151 other effect Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
Classifications
-
- 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
- G06V40/23—Recognition of whole body movements, e.g. for sport training
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
-
- 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
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Geophysics And Detection Of Objects (AREA)
Abstract
The invention discloses a sensing behavior identification method and device. The sensing behavior identification method comprises the following steps: acquiring sensing information acquired by an infrared sensing device, wherein the infrared sensing device is composed of sensor arrays arranged at different acquisition points; constructing a target information matrix according to the sensing information, wherein the target information matrix is used for representing the sensing data determined by the sensor array at different time points; and analyzing the target information matrix according to a deep convolutional neural network algorithm to determine the sensing behavior. The scheme realizes human behavior recognition with small calculated data amount, non-invasive type, low cost and small volume, and improves the recognition accuracy.
Description
Technical field
The present embodiments relate to Activity recognition technology more particularly to a kind of sensing Activity recognition method and apparatus.
Background technique
Human bodys' response technology mainly passes through physical activity in the equipment such as camera, sensor acquisition certain period of time
Data information, and pass through the intelligent recognition of algorithm realization human body behavior.Currently, the intelligence such as Human bodys' response technology and smart home
It is close industry relations to be changed.The acquisition scheme of human body behavioural information is varied, can be divided into the Image Acquisition side by camera
Case, wearable sensors acquisition scheme, passive sensor acquisition scheme.
In the prior art, the Human bodys' response based on camera information acquisition, it is desirable that multiple groups camera continuous acquisition, number
According to larger, the higher cost of amount, and consider the factor of privacy, intrusion sense is stronger in the home life.Wearable sensors not by
The influence of illumination variation, but inconvenience is dressed, user experience is poor.Passive infrared sensor is with respect to camera to illumination variation
More robustness, but it is inflexible, it is poor to more complex Activity recognition effect.Meanwhile with geometry logic identification human body row
For method because it is simple, efficiently, adaptability is good the advantages that in product market using more, but be difficult to meet relative complex
Activity recognition demand.Bayes's human body recognition method needs data volume small, theoretical perfect, is suitable for sensor human bioequivalence, but
It is poor for complex behavior in continuous time and inenarrable similar Activity recognition effect.The method of convolutional neural networks is known
Other effect is preferable, but the data dependent on acquisition, is usually applied to the computer vision Activity recognition based on camera.
Summary of the invention
The present invention provides a kind of sensing Activity recognition method and apparatus, realizes and calculates that data volume is few, non-intrusion type, cost
Human bodys' response low, small in size, improves recognition accuracy.
In a first aspect, the embodiment of the invention provides a kind of sensing Activity recognition methods characterized by comprising
It obtains through the collected heat transfer agent of infrared sensing device, the infrared sensing device in different acquisition point by setting
The sensor array composition set；
Target information matrix is constructed according to the heat transfer agent, the target information matrix is for indicating the sensor array
It is listed in the sensing data that different time points determine；
The target information matrix is analyzed according to deep layer convolutional neural networks algorithm, determines sensing behavior.
Preferably, the collection point is evenly distributed in physical space, and the sensor array is listed in each collection point
It is arranged according to preset interval, the sensor array is made of six binary system infrared sensors.
In any of the above-described scheme, it is preferred that described to include: according to heat transfer agent building target information matrix
The correction data of heat transfer agent is obtained by difference arithmetic, constructs target information matrix according to the correction data；
The target information matrix is three-dimensional matrice, and the row (the 1st dimension) of the three-dimensional matrice is the corresponding sensing data of different time points,
Column (the 2nd dimension) are the different corresponding sensing datas in collection point, and the number of plies (the 3rd dimension) is that the sensor array of different height collects
Sensing data.
In any of the above-described scheme, it is preferred that the foundation deep layer convolutional neural networks algorithm is to the target information
Matrix carries out analysis
According to formula xjmax=max (xji), i=1,2 ..., r × c and xjmin=min (xji), i=1,2 ..., r × c
The maxima and minima of data in the target information matrix is determined respectively, wherein j is the number of plies of the target information matrix,
R is the line number of the target information matrix, and c is the target information matrix column number；
According to formulaThe target information matrix is subjected to matrixing, be transformed into [- 1,
1] space；
Using transformed matrix as the input data of convolutional neural networks, according to formula
Determine the numerical value of every layer of neural network, wherein l indicates the number of plies of convolutional neural networks, and j indicates the layer network convolution
The number of plies afterwards, f indicate the number of plies before the layer network convolution,For each layer input data, first layer is the target letter after conversion
Matrix is ceased, other layers are upper one layer of output data,For multiple dimensioned convolution kernel, blFor biasing, random initializtion, σ is activation
Function, T represent Chi Hua.
In any of the above-described scheme, it is preferred that according to deep layer convolutional neural networks algorithm to the target information matrix
It is analyzed, determines that sensing behavior includes:
The target information matrix is analyzed according to deep layer convolutional neural networks algorithm, according to analysis result and is preset
Sensing, which acts, determines sensing behavior.
Second aspect, the embodiment of the invention also provides a kind of sensing Activity recognition devices, comprising:
Heat transfer agent obtains module, for obtaining through the collected heat transfer agent of infrared sensing device, the infrared biography
Induction device is made of the sensor array being arranged in different acquisition point；
Target information matrix deciding module, for constructing target information matrix, the target letter according to the heat transfer agent
Breath matrix is for indicating the sensing data that the sensor array determines in different time points；
Sense behavioural analysis determining module, for according to deep layer convolutional neural networks algorithm to the target information matrix into
Row analysis, determines sensing behavior.
Preferably, the collection point is evenly distributed in physical space, and the sensor array is listed in each collection point
It is arranged according to preset interval, the sensor array is made of six binary system infrared sensors.
In any of the above-described scheme, it is preferred that the target information matrix deciding module is specifically used for:
The correction data of heat transfer agent is obtained by difference arithmetic, constructs target information matrix according to the correction data；
The target information matrix is three-dimensional matrice, and the corresponding sensing data of behavior different time points of the three-dimensional matrice is classified as not
The same corresponding sensing data in collection point, the number of plies are the collected sensing data of sensor array of different height.
In any of the above-described scheme, it is preferred that the sensing behavioural analysis determining module is specifically used for:
According to formula xjmax=max (xji), i=1,2 ..., r × c and xjmin=min (xji), i=1,2 ..., r × c
The maxima and minima of data in the target information matrix is determined respectively, wherein j is the number of plies of the target information matrix,
R is the line number of the target information matrix, and c is the target information matrix column number；
According to formulaThe target information matrix is subjected to matrixing, be transformed into [- 1,
1] space；
Using transformed matrix as the input data of convolutional neural networks, according to formula
Determine the numerical value of every layer of neural network, wherein l indicates the number of plies of convolutional neural networks, and j indicates the layer network convolution
The number of plies afterwards, f indicate the number of plies before the layer network convolution,For each layer input data, first layer is the target letter after conversion
Matrix is ceased, other layers are upper one layer of output data,For multiple dimensioned convolution kernel, blFor biasing, random initializtion, σ is activation
Function, T represent Chi Hua.
In any of the above-described scheme, it is preferred that the sensing behavioural analysis determining module is specifically used for:
The target information matrix is analyzed according to deep layer convolutional neural networks algorithm, according to analysis result and is preset
Sensing, which acts, determines sensing behavior.
The present invention is by obtaining through the collected heat transfer agent of infrared sensing device, and the infrared sensing device is not by
With the sensor array composition of collection point setting；Target information matrix, the target information square are constructed according to the heat transfer agent
Battle array is for indicating the sensing data that the sensor array determines in different time points；According to deep layer convolutional neural networks algorithm pair
The target information matrix is analyzed, and determines sensing behavior, solve in the prior art identification sensing behavior be higher cost,
The poor problem of effect realizes and calculates that data volume is few, non-intrusion type, Human bodys' response at low cost, small in size, improves
Recognition accuracy.
Detailed description of the invention
Fig. 1 is the flow chart for the sensing Activity recognition method that the embodiment of the present invention one provides；
Fig. 2 is the position view for the collection point that the embodiment of the present invention one provides；
Fig. 3 is one layer of collected heat transfer agent of sensor array in infrared sensing device provided in an embodiment of the present invention
Schematic diagram；
Fig. 4 is the convolutional neural networks schematic diagram that the embodiment of the present invention one provides；
Fig. 5 is that the convolution algorithm boundary cycle that the embodiment of the present invention one provides extends schematic diagram；
Fig. 6 is the convolutional neural networks training schematic diagram that the embodiment of the present invention one provides；
Fig. 7 is that the convolutional neural networks training single that the embodiment of the present invention one provides exports result schematic diagram；
Fig. 8 is the structure chart of sensing Activity recognition device provided by Embodiment 2 of the present invention.
Specific embodiment
The present invention is described in further detail with reference to the accompanying drawings and examples.It is understood that this place is retouched
The specific embodiment stated is used only for explaining the present invention rather than limiting the invention.It also should be noted that in order to just
Only the parts related to the present invention are shown in description, attached drawing rather than entire infrastructure.
Embodiment one
Fig. 1 is the flow chart for the sensing Activity recognition method that the embodiment of the present invention one provides, and the present embodiment is applicable to pair
The case where human body behavior is identified, this method can be executed by calculating equipment such as computer, be specifically comprised the following steps:
Step 101 is obtained through the collected heat transfer agent of infrared sensing device, and the infrared sensing device is by difference
The sensor array composition of collection point setting.
Wherein, infrared sensing device refers to measurement using infrared ray as medium, sensor-based system.In this programme it needs to be determined that
The space of sensing behavior is provided with the infrared sensing device.Illustratively, it is assumed that there are a length, width and height are all that 2 meters of physics is empty
Between, it needs the behavior occurred to the space to identify at this time, can be the human body behavior to human body existing for the space and carry out
Identification.Specifically, two collection points are arranged in each edge in the physical space, Fig. 2 is the acquisition that the embodiment of the present invention one provides
The position view of point, as shown in Fig. 2, collection point 1 is evenly distributed in physical space to collection point 8.It the collection point can basis
The difference of specific physical space and adaptability is arranged, skilled person will appreciate that, collection point obtained acquisition data more
It is more accurate.Wherein, the different height at each collection point is provided with sensor array, illustratively, is arranged every 25 centimetres
One layer of sensor array, i.e., each collection point above-mentioned are in 25 centimetres of different height interval and are disposed with sensor array.
Each sensor array is made of binary system infrared sensor, in the present solution, each sensor array may include six two into
Sensor processed.Wherein, each sensor array can identify whether detect on 8 directions by this six binary sensors
Sensing data.8 straight lines emitted at each collection point in Fig. 2 represent on 8 detectable directions.
Specifically, each sensor array identifies whether detect sensing on 8 directions by six binary sensors
Data can take following method: it is assumed that each binary sensor recognizes sensing data, then set is " 1 ", unidentified to then setting
Position is " 0 ", and since each sensor array includes six binary sensors, i.e., the data that each sensor array detects can
It is made of 6 binary numbers, such as " 000111 ", is passed wherein corresponding to first binary system in sensor array for first " 0 "
The collection result of sensor, second " 0 " correspond to the collection result of second binary sensor in sensor array, successively class
It pushes away；Before detection, precoding can be carried out for this six bit binary data, different codings corresponds to one in 8 directions
LDPC (Low Density Parity Check Code, low density parity check code) coding can be used in direction, coding mode,
Illustratively, " 00000 " represents first direction in 8 directions, and " 010010 " represents the second direction in 8 directions,
" 101110 " represent the third direction in 8 directions, and so on；When 6 determined in sensor array are binary data
When consistent with one in corresponding six bit binary data in 8 directions having in precoding, it is determined that the sensor array
Sensing data is detected in this direction, and the infrared sensing device can adopt the sensing behavior in physical space as a result,
Collection.Fig. 3 is one layer of collected heat transfer agent of sensor array in the infrared sensing device that the embodiment of the present invention one provides
Schematic diagram can then assert array 2 as shown in figure 3, illustratively, the coding of array 2 and array 5 is corresponding with the direction of precoding
Sensing data is detected on preset direction with array 5.
Step 102 constructs target information matrix according to the heat transfer agent, and the target information matrix is for indicating described
The sensing data that sensor array determines in different time points.
Wherein, target information matrix is three-dimensional matrice, the corresponding sensing number of the behavior different time points of the three-dimensional matrice
According to being classified as the corresponding sensing data in different collection points, the number of plies is the collected sensing data of sensor array of different height.
Illustratively, the time interval between different time points can be 0.3s, i.e. the element of target information matrix represents a certain moment
The human body target distributed intelligence in certain section of altitude range of a certain node.Optionally, because of limitation (such as certain section of side of specific physical space
Edge cannot install acquisition node etc.), the correction data of heat transfer agent can be obtained by difference arithmetic, according to the correction data structure
Build target information matrix.Specifically, can be, it is assumed that node A moves right distance m to point C, and node B is located at a left side of node a
Side, and B is n away from A distance, then according to the data of the available point B of data a, c at point A and C (ancestor node before not mobile)
Optionally, in order to improve the accuracy that infrared sensing device acquires data, each junction sensor array acquisition can be sought
To the sum that changes in adjacent time of 6 data, the number that a sensor array of such as a certain node acquires at a certain moment
According to for " 110010 ", it is assumed that it is " 001011 " that the sensor array, which is listed in time data, then changed under current time to have 4
Position successively counts each node (each node includes 8 sensor arrays) collected data and changes in total period
Total bit, and successively sort according to size, the high position that serial number is subtracted one as this node data, as the node of the example above becomes
The total bit of change comes the 4th, then the data of the sensor array of the moment the example above become " 011 110010 ", due to this
Data embody the timing of sensor array acquisition, reduce measurement error.
Step 103 analyzes the target information matrix according to deep layer convolutional neural networks algorithm, determines sensing row
For.
Illustratively, using 3 layers of convolutional network, every layer of convolution nuclear structure is as shown in figure 4, Fig. 4 is the embodiment of the present invention one
The convolutional neural networks schematic diagram of offer, wherein maximum convolution kernel size kernelsize=3.Fig. 5 is the embodiment of the present invention one
The convolution algorithm boundary cycle of offer extends schematic diagram, is illustrated in figure 5 the signal of convolution algorithm boundary cycle extension pad=1
As a result.Optionally, according to formula xjmax=max (xji), i=1,2 ..., r × c and xjmin=min (xji), i=1,2 ..., r
× c determines the maxima and minima of data in the target information matrix respectively, and wherein j is the layer of the target information matrix
Number, r are the line number of the target information matrix, and c is the target information matrix column number；
According to formulaThe target information matrix is subjected to matrixing, be transformed into [- 1,
1] space；
Using transformed matrix as the input data of convolutional neural networks, according to formula
Determine the numerical value of every layer of neural network, wherein l indicates the number of plies of convolutional neural networks, and j indicates the layer network convolution
The number of plies afterwards, f indicate the number of plies before the layer network convolution,For each layer input data, first layer is the target letter after conversion
Matrix is ceased, other layers are upper one layer of output data,For multiple dimensioned convolution kernel, blFor biasing, random initializtion, σ is activation
Function, T represent Chi Hua.
By to each convolutional layer calculated result h in this method(l)Interpolation realization is carried out, in the training of convolutional neural networks
In the process, training convolutional layer 1 obtains 8*8*32 matrix first, and training pool layer 1 obtains the matrix of 6*6*32, then equally using upper
Formula training convolutional layer 2 and pond layer 2 are stated, the matrix of 4*4*64 is obtained, then also passes through convolutional layer 3 and pond layer 3 obtains 2*
Then the matrix of 2*32 trains full articulamentum, full articulamentum is similar to convolutional layer, and kernelsize is changed to 1, pad=0, then adopts
Training error is sought with softmax-Loss loss function, then updates error back propagationAnd bl, iteration training
8000 times,And blIt approaches constant.Fig. 6 is the convolutional neural networks training schematic diagram that the embodiment of the present invention one provides, such as Fig. 6
It is shown, it is assumed that the 8*8 matrix on the left side to be divided by every big lattice 2*2, that can be divided into lattice A 4*4 bigij, i expression line number, j expression column
Number, then as shown in figure, according to A11、A12、A21、A22This four big lattice obtain the right upper left 4 small lattice b of matrix11、b12、
b21、b22, and so on, according to A12、A13、A22、A23Obtain b13、b14、b23、b24, until according to A33、A34、A43、A44Obtain b55、
b56、b65、b66。
In this step, the target information matrix is analyzed according to deep layer convolutional neural networks algorithm, determines sensing
Behavior includes: to analyze according to deep layer convolutional neural networks algorithm the target information matrix, according to analysis result and in advance
If sensing, which acts, determines sensing behavior.Illustratively, the setting that default sensing movement can be carried out according to the classification of physical activity, is moved
Can be tie the shoelace, sitting posture is stretched oneself, jog, polishing-shoes, sitting posture chest expanding exercise, original place jete etc..
Specifically, input data are as follows: according to formulaTransformed 8*8*8 target information square
Battle array I.
1) 32 convolution kernel K convolutional layer 1: are initialized by xavier methodj, using boundary extending method shown in fig. 5, by
Following formula obtain 32 corresponding 8*8 characteristic pattern c(1)。
Wherein, 1,2 ..., 32 j；F is 1,2 ..., 8.
2) nonlinear change Tanh layer 1: is carried out according to tanh function.
3) pond layer 1: 8*8*32 matrixing is 6*6*32 matrix by method according to Fig.6,.
4) regularization layer 1: according to formulaPond layer output data is transformed to [- 1,
1] in range, matrix n is obtained(1), x is the element in matrix.
5) 64 convolution kernel K convolutional layer 2: are initialized by xavier methodj, defeated using boundary extending method shown in fig. 5
Enter the 6*6*32 matrix exported for regularization layer 1, obtains 64 corresponding 6*6 characteristic pattern c by following formula(2)。
Wherein, 1,2 ..., 64 j；F is 1,2 ..., 32.
6) nonlinear change Tanh layer 2: is carried out according to tanh function.
7) pond layer 2: 6*6*64 matrixing is 4*4*64 matrix by method according to Fig.6,.
8) regularization layer 2: according to formulaPond layer output data is transformed to [-
1,1] in range, matrix n is obtained(2), x is the element in matrix.
9) 32 convolution kernel K convolutional layer 3: are initialized by xavier methodj, defeated using boundary extending method shown in fig. 5
Enter the 4*4*64 matrix exported for regularization layer 2, obtains 32 corresponding 4*4 characteristic pattern c by following formula(3)。
Wherein, 1,2 ..., 32 j；F is 1,2 ..., 64.
10) nonlinear change Tanh layer 3: is carried out according to tanh function.
11) pond layer 3: 4*4*64 matrixing is 2*2*32 matrix by method according to Fig.6,.
12) regularization layer 3: according to formulaPond layer output data is transformed to [-
1,1] in range, matrix n is obtained(3), x is the element in matrix.
13) output layer: the behavior classification O of outputi=∑ wx(3), i is classification number, byIt asks
The element that softmax loss function is, x(3)ForElement, w be initialization weight.
14) changing value that each layer parameter and convolution kernel are found out by way of derivation updates each layer parameter.
15) new data and label are inputted, repeat 1) to train 8000 times to 14) step, parameter is made to tend to be constant.
By above-mentioned algorithm, final output is as shown in fig. 7, Fig. 7 is the convolutional Neural that the embodiment of the present invention one provides
Network training single exports result schematic diagram, wherein abscissa is identifiable behavior in preset 6, and illustratively, 1 indicates
It ties the shoelace；2 indicate to stretch oneself；3 indicate to jog；4 indicate polishing-shoes；5 indicate chest expanding exercise；6 indicate to capriole.Correspondingly, vertical
Coordinate is the identification probability of corresponding classification, and since the classification movement similarity having is higher, a movement may be identified as different
Behavior, taking probability highest time corresponding classification at this time is the behavior finally identified, and the probability of the 2nd classification as shown in the figure is most
Height, the then behavior determined by this algorithm " stretching oneself " movement.
The technical solution of the present embodiment, it is described infrared by obtaining through the collected heat transfer agent of infrared sensing device
Sensing device is made of the sensor array being arranged in different acquisition point；Target information matrix is constructed according to the heat transfer agent,
The target information matrix is for indicating the sensing data that the sensor array determines in different time points；According to deep layer convolution
Neural network algorithm analyzes the target information matrix, determines sensing behavior, solves to sense in identification in the prior art
Behavior is the poor problem of higher cost, effect, realizes and calculates that data volume is few, non-intrusion type, human body at low cost, small in size
Activity recognition improves recognition accuracy.
Embodiment two
Fig. 8 is the structure chart of sensing Activity recognition device provided by Embodiment 2 of the present invention, is specifically included as follows:
Heat transfer agent obtains module 201, described infrared for obtaining through the collected heat transfer agent of infrared sensing device
Sensing device is made of the sensor array being arranged in different acquisition point；
Target information matrix deciding module 202, for constructing target information matrix, the target according to the heat transfer agent
Information matrix is for indicating the sensing data that the sensor array determines in different time points；
Behavioural analysis determining module 203 is sensed, for foundation deep layer convolutional neural networks algorithm to the target information square
Battle array is analyzed, and determines sensing behavior.
The technical solution of the present embodiment, it is described infrared by obtaining through the collected heat transfer agent of infrared sensing device
Sensing device is made of the sensor array being arranged in different acquisition point；Target information matrix is constructed according to the heat transfer agent,
The target information matrix is for indicating the sensing data that the sensor array determines in different time points；According to deep layer convolution
Neural network algorithm analyzes the target information matrix, determines sensing behavior, solves to sense in identification in the prior art
Behavior is the poor problem of higher cost, effect, realizes and calculates that data volume is few, non-intrusion type, human body at low cost, small in size
Activity recognition improves recognition accuracy.
Based on the above technical solution, the collection point is evenly distributed in physical space, the sensor array
It is arranged in each collection point according to preset interval, the sensor array is made of six binary system infrared sensors.
Based on the above technical solution, the target information matrix deciding module is specifically used for:
The correction data of heat transfer agent is obtained by difference arithmetic, constructs target information matrix according to the correction data；
The target information matrix is three-dimensional matrice, and the corresponding sensing data of behavior different time points of the three-dimensional matrice is classified as not
The same corresponding sensing data in collection point, the number of plies are the collected sensing data of sensor array of different height.
Based on the above technical solution, the sensing behavioural analysis determining module is specifically used for:
According to formula xjmax=max (xji), i=1,2 ..., r × c and xjmin=min (xji), i=1,2 ..., r × c
The maxima and minima of data in the target information matrix is determined respectively, wherein j is the number of plies of the target information matrix,
R is the line number of the target information matrix, and c is the target information matrix column number；
According to formulaThe target information matrix is subjected to matrixing, be transformed into [- 1,
1] space；
Using transformed matrix as the input data of convolutional neural networks, according to formula
Determine the numerical value of every layer of neural network, wherein l indicates the number of plies of convolutional neural networks, and j indicates the layer network convolution
The number of plies afterwards, f indicate the number of plies before the layer network convolution,For each layer input data, first layer is the target letter after conversion
Matrix is ceased, other layers are upper one layer of output data,For multiple dimensioned convolution kernel, blFor biasing, random initializtion, σ is activation
Function, T represent Chi Hua.
Based on the above technical solution, the sensing behavioural analysis determining module is specifically used for:
The target information matrix is analyzed according to deep layer convolutional neural networks algorithm, according to analysis result and is preset
Sensing, which acts, determines sensing behavior.
The said goods can be performed the embodiment of the present invention one provided by method, have the corresponding functional module of execution method and
Beneficial effect.
Note that the above is only a better embodiment of the present invention and the applied technical principle.It will be appreciated by those skilled in the art that
The invention is not limited to the specific embodiments described herein, be able to carry out for a person skilled in the art it is various it is apparent variation,
It readjusts and substitutes without departing from protection scope of the present invention.Therefore, although being carried out by above embodiments to the present invention
It is described in further detail, but the present invention is not limited to the above embodiments only, without departing from the inventive concept, also
It may include more other equivalent embodiments, and the scope of the invention is determined by the scope of the appended claims.
Claims (8)
1. a kind of sensing Activity recognition method characterized by comprising
It obtains through the collected heat transfer agent of infrared sensing device, the infrared sensing device in different acquisition point by being arranged
Sensor array composition；
Target information matrix is constructed according to the heat transfer agent, the target information matrix is for indicating that the sensor array is listed in
The sensing data that different time points determine；
The target information matrix is analyzed according to deep layer convolutional neural networks algorithm, determines sensing behavior；
Wherein, the foundation deep layer convolutional neural networks algorithm, which analyze to the target information matrix, includes:
According to formula xjmax=max (xji), i=1,2 ..., r × c and xjmin=min (xji), i=1,2 ..., r × c difference
Determine the maxima and minima of data in the target information matrix, wherein j is the number of plies of the target information matrix, and r is
The line number of the target information matrix, c are the target information matrix column number；
According to formulaThe target information matrix is subjected to matrixing, is transformed into [- 1,1] sky
Between；
Using transformed matrix as the input data of convolutional neural networks, according to formula
Determine the numerical value of every layer of neural network, wherein l indicates the number of plies of convolutional neural networks, after j indicates the layer network convolution
The number of plies, f indicate the number of plies before the layer network convolution,For each layer input data, first layer is the target information square after conversion
Battle array, other layers are upper one layer of output data,For multiple dimensioned convolution kernel, blFor biasing, random initializtion, σ is activation letter
Number, T represent Chi Hua.
2. described the method according to claim 1, wherein the collection point is evenly distributed in physical space
Sensor array is listed in each collection point to be arranged according to preset interval, and the sensor array is by six binary system infrared sensors
Composition.
3. the method according to claim 1, wherein described construct target information matrix according to the heat transfer agent
Include:
The correction data of heat transfer agent is obtained by difference arithmetic, constructs target information matrix according to the correction data；It is described
Target information matrix is three-dimensional matrice, and the corresponding sensing data of behavior different time points of the three-dimensional matrice is classified as different
The corresponding sensing data in collection point, the number of plies are the collected sensing data of sensor array of different height.
4. the method according to claim 1, wherein believing according to deep layer convolutional neural networks algorithm the target
Breath matrix is analyzed, and determines that sensing behavior includes:
The target information matrix is analyzed according to deep layer convolutional neural networks algorithm, according to analysis result and default sensing
It acts and determines sensing behavior.
5. a kind of sensing Activity recognition device characterized by comprising
Heat transfer agent obtains module, for obtaining through the collected heat transfer agent of infrared sensing device, the infrared sensing dress
It sets and is made of the sensor array being arranged in different acquisition point；
Target information matrix deciding module, for constructing target information matrix, the target information square according to the heat transfer agent
Battle array is for indicating the sensing data that the sensor array determines in different time points；
Behavioural analysis determining module is sensed, for dividing according to deep layer convolutional neural networks algorithm the target information matrix
Analysis, determines sensing behavior；
Wherein, the sensing behavioural analysis determining module is specifically used for:
According to formula xjmax=max (xji), i=1,2 ..., r × c and xjmin=min (xji), i=1,2 ..., r × c difference
Determine the maxima and minima of data in the target information matrix, wherein j is the number of plies of the target information matrix, and r is
The line number of the target information matrix, c are the target information matrix column number；
According to formulaThe target information matrix is subjected to matrixing, is transformed into [- 1,1] sky
Between；
Using transformed matrix as the input data of convolutional neural networks, according to formula
Determine the numerical value of every layer of neural network, wherein l indicates the number of plies of convolutional neural networks, after j indicates the layer network convolution
The number of plies, f indicate the number of plies before the layer network convolution,For each layer input data, first layer is the target information square after conversion
Battle array, other layers are upper one layer of output data,For multiple dimensioned convolution kernel, blFor biasing, random initializtion, σ is activation letter
Number, T represent Chi Hua.
6. device according to claim 5, which is characterized in that the collection point is evenly distributed in physical space, described
Sensor array is listed in each collection point to be arranged according to preset interval, and the sensor array is by six binary system infrared sensors
Composition.
7. device according to claim 5, which is characterized in that the target information matrix deciding module is specifically used for:
The correction data of heat transfer agent is obtained by difference arithmetic, constructs target information matrix according to the correction data；It is described
Target information matrix is three-dimensional matrice, and the corresponding sensing data of behavior different time points of the three-dimensional matrice is classified as different
The corresponding sensing data in collection point, the number of plies are the collected sensing data of sensor array of different height.
8. device according to claim 5, which is characterized in that the sensing behavioural analysis determining module is specifically used for:
The target information matrix is analyzed according to deep layer convolutional neural networks algorithm, according to analysis result and default sensing
It acts and determines sensing behavior.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611024234.5A CN106446876B (en) | 2016-11-17 | 2016-11-17 | Sensing behavior identification method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611024234.5A CN106446876B (en) | 2016-11-17 | 2016-11-17 | Sensing behavior identification method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106446876A CN106446876A (en) | 2017-02-22 |
| CN106446876B true CN106446876B (en) | 2019-10-18 |
Family
ID=58220537
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611024234.5A Active CN106446876B (en) | 2016-11-17 | 2016-11-17 | Sensing behavior identification method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106446876B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107966447B (en) * | 2017-11-14 | 2019-12-17 | 浙江大学 | workpiece surface defect detection method based on convolutional neural network |
| CN108764059B (en) * | 2018-05-04 | 2021-01-01 | 南京邮电大学 | Human behavior recognition method and system based on neural network |
| CN109009013A (en) * | 2018-06-25 | 2018-12-18 | 彭礼烨 | Intelligent bracelet |
| CN109768944A (en) * | 2018-12-29 | 2019-05-17 | 苏州联讯仪器有限公司 | A kind of signal modulation identification of code type method based on convolutional neural networks |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105353876A (en) * | 2015-11-09 | 2016-02-24 | 深圳市云腾智慧科技有限公司 | Multi-point light sensation based spatial gesture identification control system and method |
| CN105868779A (en) * | 2016-03-28 | 2016-08-17 | 浙江工业大学 | Method for identifying behavior based on feature enhancement and decision fusion |
-
2016
- 2016-11-17 CN CN201611024234.5A patent/CN106446876B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105353876A (en) * | 2015-11-09 | 2016-02-24 | 深圳市云腾智慧科技有限公司 | Multi-point light sensation based spatial gesture identification control system and method |
| CN105868779A (en) * | 2016-03-28 | 2016-08-17 | 浙江工业大学 | Method for identifying behavior based on feature enhancement and decision fusion |
Non-Patent Citations (2)
| Title |
|---|
| 一种基于三维加速度传感器的人体行为识别方法;徐川龙，顾勤龙，姚明海;《计算机系统应用》;20131231;第22卷(第6期);132-135 * |
| 时空域深度卷积神经网络及其在行为识别上的应用;刘琮，许维胜，吴启迪;《计算机科学》;20150731;第42卷(第7期);P245-249 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106446876A (en) | 2017-02-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Phyo et al. | Deep learning for recognizing human activities using motions of skeletal joints | |
| CN106446876B (en) | Sensing behavior identification method and device | |
| Haque et al. | Towards viewpoint invariant 3d human pose estimation | |
| CN111310707B (en) | Bone-based graph annotation meaning network action recognition method and system | |
| CN107818314B (en) | Face image processing method, device and server | |
| EP3180741B1 (en) | Multi-layer aggregation for object detection | |
| Tosi et al. | Metrics for probabilistic geometries | |
| AU2019374875B2 (en) | Identifying image aesthetics using region composition graphs | |
| CN106682697A (en) | End-to-end object detection method based on convolutional neural network | |
| CN102360494B (en) | Interactive image segmentation method for multiple foreground targets | |
| Samanta et al. | Space-time facet model for human activity classification | |
| Wang et al. | RAR-U-Net: a residual encoder to attention decoder by residual connections framework for spine segmentation under noisy labels | |
| Nishi et al. | Generation of human depth images with body part labels for complex human pose recognition | |
| CN105469041A (en) | Facial point detection system based on multi-task regularization and layer-by-layer supervision neural networ | |
| CN106548194B (en) | The construction method and localization method of two dimensional image human joint points location model | |
| Guan et al. | Compressive classification of human motion using pyroelectric infrared sensors | |
| CN104463148B (en) | Face identification method based on Image Reconstruction and hash algorithm | |
| CN112487948B (en) | Multi-space fusion-based concentration perception method for learner in learning process | |
| CN106056135A (en) | Human body motion classification method based on compression perception | |
| Fan et al. | Efficient isometric multi-manifold learning based on the self-organizing method | |
| Jain et al. | Ambient intelligence-based multimodal human action recognition for autonomous systems | |
| CN115713805A (en) | Body-building action recognition monitoring method and system | |
| Wang | Three-dimensional convolutional restricted Boltzmann machine for human behavior recognition from RGB-D video | |
| Fu et al. | Sparse coding-based space-time video representation for action recognition | |
| CN114241270A (en) | Intelligent monitoring method, system and device for home care |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |