CN104599290B - Video sensing node-oriented target detection method - Google Patents
Video sensing node-oriented target detection method Download PDFInfo
- Publication number
- CN104599290B CN104599290B CN201510023864.XA CN201510023864A CN104599290B CN 104599290 B CN104599290 B CN 104599290B CN 201510023864 A CN201510023864 A CN 201510023864A CN 104599290 B CN104599290 B CN 104599290B
- Authority
- CN
- China
- Prior art keywords
- background
- pixel
- image
- matrix
- target area
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 44
- 238000009826 distribution Methods 0.000 claims abstract description 21
- 238000005070 sampling Methods 0.000 claims abstract description 11
- 230000008859 change Effects 0.000 claims abstract description 7
- 238000012805 post-processing Methods 0.000 claims abstract description 5
- 239000011159 matrix material Substances 0.000 claims description 62
- 238000000034 method Methods 0.000 claims description 33
- 238000005259 measurement Methods 0.000 claims description 19
- 230000008569 process Effects 0.000 claims description 13
- 239000013598 vector Substances 0.000 claims description 10
- 230000003044 adaptive effect Effects 0.000 claims description 9
- 238000006243 chemical reaction Methods 0.000 claims description 5
- 230000006835 compression Effects 0.000 claims description 5
- 238000007906 compression Methods 0.000 claims description 5
- 239000004744 fabric Substances 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 description 17
- 238000010586 diagram Methods 0.000 description 5
- 230000036961 partial effect Effects 0.000 description 5
- 230000008901 benefit Effects 0.000 description 4
- 238000012544 monitoring process Methods 0.000 description 4
- 230000007812 deficiency Effects 0.000 description 3
- 239000000203 mixture Substances 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000033001 locomotion Effects 0.000 description 2
- 230000008447 perception Effects 0.000 description 2
- 230000002829 reductive effect Effects 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 238000011410 subtraction method Methods 0.000 description 2
- 235000008331 Pinus X rigitaeda Nutrition 0.000 description 1
- 235000011613 Pinus brutia Nutrition 0.000 description 1
- 241000018646 Pinus brutia Species 0.000 description 1
- 206010044565 Tremor Diseases 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000002902 bimodal effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000003628 erosive effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000002156 mixing Methods 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a video sensing node-oriented target detection method. The video sensing node-oriented target detection method includes steps of compressed sensing, background modeling, updating and post-processing, different sampling values M are set according to interest regions, the sampling rate is improved at the 1.2 times region of the previous frame of a target block, and the sampling rate is lowered at the background region; when the background brightness change is small, the Gaussian distribution number for modeling is lowered to lower the learning speed rate; when the brightness change is large, the Gaussian distribution number is improved to improve the learning speed rate.
Description
Technical field
The present invention relates to target detection technique field, more particularly to a kind of target detection side towards video-aware node
Method.
Background technology
Wireless video sensor network by a large amount of video nodes with communication and computing capability in a specific manner or
It is randomly disposed as " intelligence " autonomy observing and controlling Radio Network System constituted in monitor area.Have very between video sensor node
Strong cooperative ability, by the data interaction between the image data acquiring of local, process and node global task is completed.With biography
System monitoring mode is compared, and is built distributed intelligent monitoring system using wireless video sensor network and is had unmanned, covers
Rate is wide, stable performance, flexibility are high, monitoring scene can realize the advantage that is combined, is particularly suitable on traffic intersection, airport
With the key area such as subway station or the target following under adverse circumstances and event monitoring.
In computer vision and wireless video sensor network related application field, to the target in video image that obtains
Detection is primary step.The quality of algorithm of target detection is had influence on at the further vision such as follow-up tracking and Activity recognition
Reason.Complicated and changeable due to actual scene causes existing algorithm of target detection generally more complicated, computationally intensive, memory size
Have high demands, be not suitable for the video-aware node of resource-constrained.
Therefore, being directed to the algorithm of target detection of video-aware node must first consider the problem of algorithm efficiency, use up
It is likely to reduced amount of calculation and memory capacity.Compressive sensing theory breaches the requirement under tradition draws Qwest theoretical to sample number.
As long as signal is compressible or sparse, it is possible to believed the higher-dimension after conversion by meeting the observing matrix of certain condition
Number sampled, obtained the low-dimensional signal after a sampling.Then solving an optimization problem just can be from a small amount of sampled value
In perfectly reconstruct primary signal.Compressive sensing theory is applied in the algorithm of target detection based on background subtraction method,
While retaining original image information, the pixel quantity for participating in background modeling can be greatly decreased, so as to improve efficiency of algorithm.Cause
This, on the basis of research nowadays several conventional background modeling methods, proposes a kind of based on the adaptive of structuring compressed sensing
Answer mixed Gaussian (Structured Compressive Sensing Adaptive Gaussian Mixture Model,
SCS-AGMM) background modeling algorithm, builds structuring random measurement matrix to reduce the data volume for participating in background modeling, and from many
Individual aspect optimizes the efficiency of algorithm.
Background subtraction method is a kind of method ripe in object detection field Technical comparing, using quite varied.The party
By subtracting each other to video image present frame and background model correspondence position pixel value, the absolute values being on duty are more than certain threshold value to method
When, the pixel is judged as object pixel, it is otherwise background pixel.And by later image process, obtain complete target image.
For background more complicated and that dynamic change is presented, such as exist in scene the fluctuation water surface, the trees of shake,
Camera trembles, and the probability density distribution figure of pixel value is often presented bimodal or multimodal state.This is accomplished by using many
The linear combination of individual Gaussian Profile could be to background accurate modeling, and the method is referred to as mixed Gauss model (GMM).Using GMM pair
Situations such as each pixel in image sets up background model and adapts to the interference of illumination variation in video image, movement background.
Occur in recent years being based on mixed Gaussian background modeling innovatory algorithm in a large number, the advantage of these methods is Detection results
Preferably, the motion artifacts in the case of complex background can be removed；Deficiency is that amount of calculation and amount of storage are larger, the speed of service compared with
Slowly, it is unsuitable for the video-aware node of resource-constrained.
The content of the invention
The present invention is directed to deficiencies of the prior art, there is provided a kind of target detection towards video-aware node
Method.The present invention is achieved through the following technical solutions：
A kind of object detection method towards video-aware node, including step：
Image reconstruction step：Picture size size according to collecting carries out piecemeal to image, will adopt the image block for obtaining
Be converted to the vector of N × 1；
Compressed sensing step：Build structuring random measurement matrix carries out Sampling Compression to the vector after conversion；
Background modeling step：Gauss is carried out using ADAPTIVE MIXED Gauss model to the matrix-block after each measurement to build
Mould, using minimum pixel rule object block and background block detection are carried out；By each picture of the mixed Gauss model to each both candidate nodes
Element sets up at least one background model, and background model is initialized with the first frame image data, and each background model is set
Fixed unified background threshold, pixel weights are background distributions more than the background model description of the background threshold, and pixel is weighed
Value is distributed less than or equal to what the background model of the background threshold was described for prospect, with the pixel for being judged as background in image again just
Background model of the beginningization weights less than initial threshold value；The distributed constant of background model according to priority from big to small with it is corresponding work as
Preceding pixel value matching detection one by one, judges background model with the unmatched pixel of current pixel value as in target area
Point, to the background model that the match is successful distributed constant is updated, and to each background model weight is updated；
Update step：Object block and background block are updated using different strategies, and according to the result for detecting to knot
Structure random measurement matrix carries out parameter regulation；
Post-processing step：Target image to detecting carries out later stage process and obtains final target image；
Wherein, different strategies includes arranging different sampled values M according to interest region, in the object block 1.2 of former frame
Sample rate is improved in times region, and reduces sample rate in background area；And, when background luminance changes less, reduce the height of modeling
This distribution number, to reduce learning rate；When brightness is changed greatly, Gaussian Profile number is improved, to improve learning rate.
Preferably, distributed constant according to priority carries out matching detection one by one with current pixel value, that is, discriminate whether to meet |
μI, t-xt| ＜ max(WσI, t, τ), i=1 in formula, 2 ..., K, K are the number of each pixel Gaussian Profile, μI, tAnd σI, tRespectively in t
The average and standard variance of i-th Gaussian Profile of moment, xtFor preceding pixel value, W and τ is threshold value constant.
Preferably, the target area that previous frame is detected is matched after extension as the target area of present frame
Detection, the pixel outside target area takes higher value using tight matching criterior, i.e. τ and W；Pixel in target area
Point takes smaller value using loose matching criterior, i.e. τ and W, wherein, 0.5 ＜=W ＜=3.5,3 ＜=τ ＜=20.
Preferably, using previous frame target area extension 10% as present frame target area, the picture outside target area
Vegetarian refreshments takes W=2.5, and τ=15, the pixel in target area takes W=1.5, τ=6.
Preferably, during at least one background model of initialization, being used for initializing Gauss point by the first frame each point pixel value
Cloth mean μK, 0, the standard variance σ of the first frame each point pixel valueK, 0Take 15 ＜=σK, 0＜=25, weight is 1/Kmax, and Kmax is
The largest Gaussian one distribution number of each pixel.
The method that the present invention takes builds structuring random measurement matrix by carrying out Sampling Compression to image, reduces height
The calculating data volume of this statistical modeling, and the efficiency optimization of two aspects is carried out to algorithm.One is the change according to background luminance
Come self-adaptative adjustment Gauss model number and learning rate, reduce average calculation times；Two is to extract target according to segmentation
Interest region adopts different measuring value, overall to reduce the number of pixels for participating in modeling, effectively reduces the time of background modeling.
The results show surveyed by algorithm simulating and node, the method can obtain preferable object detection results and with compared with
Strong anti-interference, relative to traditional mixed Gaussian algorithm, memory size reduces about 3/4ths, and process time can be reduced
More than 50%.
Description of the drawings
Shown in Fig. 1 is the flow chart of the present invention；
Shown in Fig. 2 is the every frame average handling time comparison schematic diagram from different modeling methods of the invention；
Shown in Fig. 3 is the false drop rate comparison schematic diagram from different modeling methods of the invention；
Shown in Fig. 4 is the loss comparison schematic diagram from different modeling methods of the invention；
Shown in Fig. 5 is the performance from different modeling methods of the invention and average every frame process time comparison schematic diagram；
Shown in Fig. 6 is the process time from different modeling methods of the invention and memory size comparison schematic diagram.
Specific embodiment
Below with reference to the accompanying drawing of the present invention, clear, complete description is carried out to the technical scheme in the embodiment of the present invention
And discussion, it is clear that as described herein is only a part of example of the present invention, is not whole examples, based on the present invention
In embodiment, the every other enforcement that those of ordinary skill in the art are obtained on the premise of creative work is not made
Example, belongs to protection scope of the present invention.
For the ease of the understanding to the embodiment of the present invention, make by taking specific embodiment as an example further below in conjunction with accompanying drawing
Illustrate, and each embodiment does not constitute the restriction to the embodiment of the present invention.
For complicated dynamic background, accurately building to background can be realized using the linear combination of multiple Gaussian Profiles
Mould.The probability density characteristicses of each pixel in image are expressed as K (general value 3～5) Gaussian mode by mixed Gauss model
Type.Each Gaussian Profile has different weights ω i, t (∑ ω i, t=1, i=1,2 ... ..., K) and priority (ω/σ), and presses
It is ranked up from high to low according to priority.It is determined that during background distributions, certain threshold value T (T ＜ 1) is taken to single distribution, when this point
When the weights of cloth are more than or equal to the threshold value, it is believed that this Gauss model is background distributions, otherwise it is assumed that this Gaussian mode
Type is prospect distribution.
If xt is a certain pixel value of t, its probability density letter can be described with the linear combination of K Gaussian Profile
Number：
Wherein ω i, t, μ i, t and ∑ i, t are respectively in the weights of i-th Gaussian Profile of t, average and covariance
Matrix.It is assumed that each pixel color component is separate, its covariance matrix is represented by：
K Gaussian Profile presses the arrangement of ω/σ descendings, and weighting value represents background distributions more than the Gaussian Profile of a certain threshold value,
I.e.：
fi(x | μ, σ2) ∈ Bg, if ωi＞ TH (3)
Wherein, TH is background threshold.
1st, initialization background method
The image sequence for generally being collected by video node, within a period of time, the change of background be it is little,
It can thus be assumed that each pixel gray value obeys the Gaussian Profile of mean μ and standard variance σ, and the Gauss of each picture element
Distribution is independent.First initial background, is to reduce computation complexity and memory capacity, using 3 Gauss models to each
Pixel is modeled (K=3).Gaussian Profile expects that μ is initialized with the pixel value of each point in the first two field picture, standard side
Difference σ takes larger value (σ=20), weights be initially set to 1/ (2i+1) (i=0,1,2).
2nd, the study of background model and update method
According to priority order ω/σ xt is matched one by one with each Gaussian Profile from big to small when target detection is carried out.
If being not detected by representing that the Gaussian Profile of background distributions is matched with xt, then it is assumed that it is otherwise background that the point is target.Background
The concrete execution step of model algorithm is as follows：
(1) matching criterior
K Gaussian Profile is according to priority compared with current pixel value xt and sees whether meet condition | μI, t-xt| ＜
Max (W* σ, λ), (i=1,2 ... ..., K), W and λ are coefficients in formula.The pixel of target area and nontarget area is adopted not
The judgement of target and background is carried out with condition.Concrete operations are that the size of the target frame that will be detected in previous frame image is amplified
1.1 times of target frames as present frame.
Judged using following condition belonging to target area pixel：
|x-μi| ＜ max (1.5* σ, 6) (4)
The Rule of judgment of nontarget area pixel is：
|x-μi| ＜ max (2.5* σ, 15) (5)
Different value strategies is adopted according further to the position of pixel：Pixel outside for target area, takes W=2.5,
λ=15；For the pixel in target area, W=1.5, λ=6 are taken.So, the pixel in target area is easier to be detected as
Target, increased the integrity of shape of target detection.
(2) Background learning and renewal
Matched with known i-th Gauss model with the pixel value for obtaining, if the match is successful, by formula (6) renewal
The i-th Gauss model distributed constant matched somebody with somebody：
In equation (2.16)Wherein β is used for controlling the speed of background and prospect renewal (according to mould
Whether type describes background and takes different values).In most cases prospect updates slower than background.The power of K Gaussian Profile
Value updates as the following formula：
ωI, t+1=(1- αi)ωI, t+αiMI, t (7)
In equation (7)The size of α determines its priority in the background and determines each Gauss
The renewal speed of composition weights, the less background images of α are more stable；β sizes determine the renewal speed of background, and β gets over overall background image
Convergence rate is faster.
Adaptive GMM (AGMM) algorithm steps are as follows：
1) Gaussian Profile initialization (weights, expectation, variance), k=0 are carried out with the first two field picture；
2) for the new pixel of t
Matching is judged whether according to formula (4) (5)；
Then execution step 3), otherwise execution step 4)；
3) to the Gauss model for matching, it is updated using formula (6) (7)；
If 4) mismatched, with currency new Gauss model (little weights, big variance) is initialized；
5) weights variance ratio ω/σ is calculated, minimum of a value is replaced in descending arrangement；
6) basis | XT+1, i-BI, t| ＜ T judge that pixel is prospect or background, output result；
Go to step 2)；
7) terminate, next two field picture.
3rd, post-processing approach
Bianry image template Morg of target is obtained according to method described above.3 × 3 shapes are carried out to two-value template Morg
State opening operation, it is Ms to obtain result, then obtains result for M after 3 × 3 erosion operations remove isolated point.The process is led
The loss of partial target pixel is caused, has taken the following processing method based on morphology object reconstruction as far as possible to retain more
Target image：
F is the final result after foreground extraction, noise filtering in equation (8).The chi of the structural element SE in equation
Very little size depends on the target size of detection.Experiment finds that using 3 × 3 structural element preferable target detection knot can be reached
Really.Cavity filling is carried out to the foreground target F being partitioned into reference to Assimilation filling using structural element can make target more complete.Most
The result for being counted by target sizes afterwards removes the fritter less than 40 pixels, to reach the purpose for eliminating noise.
Mixed Gauss model parameter is constantly updated to adapt to gradually changing for background.In addition, the algorithm is due to figure
Each pixel carries out 3 to 5 Gaussian modelings as in, and overall calculation amount and memory capacity are larger.
4th, structuring compressed sensing algorithm
At present, conventional modeling method is to describe dynamic background using mixture Gaussian background model.Due to each pixel
3 to 5 Gauss models, the substantial amounts of calculating of method consumption shooting head node of mixed Gauss model and storage resource are set up, is affected
The real-time application of algorithm.In order to improve the efficiency of algorithm, introduce compressed sensing algorithm carries out stochastical sampling to view data, from
And reduce the amount of calculation and amount of storage of background modeling algorithm.But the completely random characteristic of stochastical sampling matrix causes hardware circuit
Implement more complicated and object detection results and there is uncertainty.For such case, the present invention compresses structuring
Perception algorithm is incorporated in the middle of Gaussian modeling, studies a kind of using structuring random measurement matrix on this basis, to figure
As the ADAPTIVE MIXED Gaussian Background modeling method sampled, and global efficiency optimization is carried out to algorithm, improve overall operation
Efficiency.
Compressed sensing is that the calculation matrix Φ for arranging (M ＜ ＜ N) size with M rows N is measured to signal x (N-dimensional), is obtained
Measured value y (M dimensions) after compression, the process can be realized by equation (9).
Y=φ x=φ Ψ α=Θ α (9)
If signal x have in certain domain of variation it is openness, such as shown in equation (10)：
α=ΨTx (10)
And calculation matrix Φ meet the constraint isometry conditions, that is, refer to for arbitrary K sparse signals f and constant δk∈
(0,1) meet：So just can be by equation (11) come perfect recovery signal：
The process is referred to as reconstructed, and what 0 norm therein referred to is exactly the number of 0 element.
The calculation matrix of the meet the constraint isometry condition for proposing at present mainly divides three classes.The first kind includes that matrix element is only
On the spot obey gaussian random calculation matrix, bernoulli random matrix of a certain distribution etc..Equations of The Second Kind includes partial orthogonality matrix, portion
Divide hadamard matrix and irrelevant calculation matrix.This matroid is only uncorrelated to the signal in time domain or frequency-domain sparse.3rd class
Including Teoplitz (Toeplitz) matrix, structuring random measurement matrix, Chirps calculation matrix, circular matrix, random volume
The perception matrix that product is formed.
1) random Gaussian matrix：As shown in formula (12), calculation matrix each element independently obeys average for 0, and variance is
The Gaussian Profile of 1/M, equiprobability value is 1 or 0.The advantage of Gauss measurement matrix be needed for measurement line number it is less and also it
It is almost all uncorrelated to any sparse signal.
2) random bernoulli matrix；As shown in formula (13), each element of calculation matrix independently obeys symmetrical Bei Nu
Profit distribution, equiprobability value is 1 or -1.The matrix randomness is very strong, with the property similar with Gaussian matrix.
3) partial orthogonality matrix；The step of building the matrix is the orthogonal matrix U for firstly generating N × N, then in matrix U
In randomly choose M row vectors and the column vector to M × N matrix carry out it is unitization, you can obtain partial orthogonality matrix.
4) Toeplitz matrixes；The step of building the matrix is to firstly generate calculation matrix Φ, matrix Φ each
In row vector, according to the probability distribution randomly chosen position of equation (14) element, then corresponding position assignment 0,1 ,-
1, wherein
5) part hadamard matrix；The step of building the matrix is to firstly generate the hadamard matrix that size is N × N, so
Randomly choose the Hadamard calculation matrix that M row vectors may make up a M × N in generator matrix afterwards.
6) structuring random measurement matrix；Although random Gaussian and random bernoulli matrix have non-to many sparse signals
Correlation, but because the characteristic of its completely random causes to calculate more complicated, memory size has high demands.Therefore many is researched and proposed
The concept of structuring random measurement matrix.The structure of this matroid adopts random Gaussian, Bernoulli Jacob's matrix and partial Fourier
The mixed model of transformation matrix, randomly selects M rows from the hybrid matrix of N × N, then each row are normalized.Knot
Structure random measurement matrix is almost uncorrelated to every other orthogonal matrix, and maintains the advantage of various matrixes.
7) certainty matrix；Completely random matrix has the shortcomings that uncertain factor and hardware circuit are difficult to, and is gram
Its deficiency in compressed sensing application is taken, many has researched and proposed certainty calculation matrix, including multinomial certainty matrix
With rotation calculation matrix etc..
The present invention includes step：
Image reconstruction step：Picture size size according to collecting carries out piecemeal to image, will adopt the image block for obtaining
Be converted to the vector of N × 1；
Compressed sensing step：Build structuring random measurement matrix carries out Sampling Compression to the vector after conversion；
Background modeling step：Gauss is carried out using ADAPTIVE MIXED Gauss model to the matrix-block after each measurement to build
Mould, using minimum pixel rule object block and background block detection are carried out；By each picture of the mixed Gauss model to each both candidate nodes
Element sets up at least one background model, and background model is initialized with the first frame image data, and each background model is set
Fixed unified background threshold, pixel weights are background distributions more than the background model description of the background threshold, and pixel is weighed
Value is distributed less than or equal to what the background model of the background threshold was described for prospect, with the pixel for being judged as background in image again just
Background model of the beginningization weights less than initial threshold value；The distributed constant of background model according to priority from big to small with it is corresponding work as
Preceding pixel value matching detection one by one, judges background model with the unmatched pixel of current pixel value as in target area
Point, to the background model that the match is successful distributed constant is updated, and to each background model weight is updated；
Update step：Object block and background block are updated using different strategies, and according to the result for detecting to knot
Structure random measurement matrix carries out parameter regulation；
Post-processing step：Target image to detecting carries out later stage process and obtains final target image；
Wherein, different strategies includes arranging different sampled values M according to interest region, in the object block 1.2 of former frame
Sample rate is improved in times region, and reduces sample rate in background area；And, when background luminance changes less, reduce the height of modeling
This distribution number, to reduce learning rate；When brightness is changed greatly, Gaussian Profile number is improved, to improve learning rate.
Distributed constant according to priority carries out matching detection one by one with current pixel value, that is, discriminate whether to meet | μI, t-xt| ＜
max(WσI, t, τ), i=1 in formula, 2 ..., K, K are the number of each pixel Gaussian Profile, μI, tAnd σI, tRespectively in t i-th
The average and standard variance of individual Gaussian Profile, xtFor preceding pixel value, W and τ is threshold value constant.
The target area that previous frame is detected carries out matching detection after extension as the target area of present frame,
Pixel outside target area takes higher value using tight matching criterior, i.e. τ and W；Pixel in target area is adopted
The matching criterior of pine, i.e. τ and W takes smaller value, wherein, 0.5 ＜=W ＜=3.5,3 ＜=τ ＜=20.By previous frame target
Used as the target area of present frame, the pixel outside target area takes W=2.5, τ=15, in target area for region extension 10%
Pixel in domain takes W=1.5, τ=6.
When initializing at least one background model, it is used for initializing Gaussian Profile average by the first frame each point pixel value
μK, 0, the standard variance σ of the first frame each point pixel valueK, 0Take 15 ＜=σK, 0＜=25, weight is 1/Kmax, and Kmax is each picture
The largest Gaussian one distribution number of vegetarian refreshments.
The image xt that the algorithm of target detection of the present invention is collected first to video node carries out 4 × 4 or 8 × 8 piecemeals, so
Afterwards build structuring random measurement matrix Φ in spatial domain directly to image sampling after obtain compress image yt.By compressed sensing
Theory understands that yt contains original image overwhelming majority information, and by ADAPTIVE MIXED Gauss model (AGMM) background mould is built
Type, by background subtraction foreground image is obtained, and then carries out Morphological scale-space to foreground image.
The application first selects activation node to carry out target detection, target following, then selects to work as by Efficiency Function f (i)
Front optimum node carries out target following, and process as shown in Figure 1, target detection is built using adaptive Gauss mixing background
Mould, realizes the detection and segmentation of moving target；The target following of node is realized by distributed Mean shift and target association
Determine sensor network measures of effectiveness function with the factor such as state estimation, Detection results, communication energy consumption with reference to sensor node,
Optimal sensor node is selected to carry out target following.Consider computation complexity, the transmission of data, storage demand, realize
Accurate tracking to moving target under interior complex scene on a large scale.As shown in Figures 2 to 6, according to the ratio with existing other algorithms
Compared with the present invention can obtain preferable object detection results and with stronger anti-interference. relative to traditional mixed Gaussian
Algorithm, memory size reduces about 3/4ths, and process time can reduce more than 50%.
The above, the only present invention preferably specific embodiment, but protection scope of the present invention is not limited thereto,
Any those familiar with the art the invention discloses technical scope in, the change or replacement that can be readily occurred in,
All should be included within the scope of the present invention.Therefore, protection scope of the present invention should be with scope of the claims
It is defined.
Claims (5)
1. a kind of object detection method towards video-aware node, it is characterised in that including step：
Image reconstruction step：Picture size size according to collecting carries out piecemeal to image, will adopt the image block conversion for obtaining
For the vector of N × 1；
Compressed sensing step：Build structuring random measurement matrix carries out Sampling Compression to the vector after conversion；
Background modeling step：Gauss modeling is carried out to the matrix-block after each measurement using ADAPTIVE MIXED Gauss model, is adopted
Object block and background block detection are carried out with minimum pixel rule；Each pixel of each both candidate nodes is set up by mixed Gauss model
At least one background model, is initialized with the first frame image data to the background model, to each background model setting
Unified background threshold, pixel weights are background distributions, pixel weights more than the background model description of the background threshold
It is distributed for prospect less than or equal to what the background model of the background threshold was described, it is again initial with the pixel for being judged as background in image
Change background model of the weights less than initial threshold value；The distributed constant of background model is according to priority current with corresponding from big to small
Pixel value matching detection one by one, judge background model with the unmatched pixel of current pixel value as the point in target area,
Distributed constant is updated to the background model that the match is successful, weight is updated to each background model；
Update step：Object block and background block are updated using different strategies, and according to the result for detecting to structuring
Random measurement matrix carries out parameter regulation；
Post-processing step：Target image to detecting carries out later stage process and obtains final target image；
Wherein, different strategies includes arranging different sampled values M according to interest region, in 1.2 times of areas of object block of former frame
Sample rate is improved in domain, and reduces sample rate in background area；And, when background luminance changes less, the Gauss point for reducing modeling
Cloth number, to reduce learning rate；When brightness is changed greatly, Gaussian Profile number is improved, to improve learning rate.
2. the object detection method towards video-aware node according to claim 1, it is characterised in that distributed constant is pressed
Priority carries out matching detection one by one with current pixel value, that is, discriminate whether to meet | μi,t-xt| ＜ max (W σi,t, τ), i in formula
=1,2 ..., K, K are the number of each pixel Gaussian Profile, μi,tAnd σi,tRespectively i-th Gaussian Profile of t average and
Standard variance, xtFor preceding pixel value, W and τ is threshold value constant.
3. the object detection method towards video-aware node according to claim 2, it is characterised in that examine previous frame
The target area measured carries out matching detection after extension as the target area of present frame, the pixel outside target area
Higher value is taken using tight matching criterior, i.e. τ and W；Pixel in target area is using loose matching criterior, i.e. τ and W
Smaller value is taken, wherein, 0.5<=W<=3.5,3<=τ<=20.
4. the object detection method towards video-aware node according to claim 3, it is characterised in that by previous frame mesh
Region extension 10% is marked as the target area of present frame, the pixel outside target area takes W=2.5, τ=15, in target
Pixel in region takes W=1.5, τ=6.
5. the object detection method towards video-aware node according to claim 1, it is characterised in that initialization is at least
During one background model, it is used for initializing Gaussian Profile mean μ by the first frame each point pixel valueK,0, the first frame each point pixel value
Standard variance σK,0Take 15<=σK,0<=25, weight is 1/Kmax, and Kmax is the largest Gaussian one distribution number of each pixel.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510023864.XA CN104599290B (en) | 2015-01-19 | 2015-01-19 | Video sensing node-oriented target detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510023864.XA CN104599290B (en) | 2015-01-19 | 2015-01-19 | Video sensing node-oriented target detection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104599290A CN104599290A (en) | 2015-05-06 |
| CN104599290B true CN104599290B (en) | 2017-05-10 |
Family
ID=53125040
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510023864.XA Active CN104599290B (en) | 2015-01-19 | 2015-01-19 | Video sensing node-oriented target detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104599290B (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105095898B (en) * | 2015-09-06 | 2019-03-08 | 苏州经贸职业技术学院 | A kind of targeted compression cognitive method towards real-time vision system |
| CN105551014A (en) * | 2015-11-27 | 2016-05-04 | 江南大学 | Image sequence change detection method based on belief propagation algorithm with time-space joint information |
| CN105761215B (en) * | 2016-01-27 | 2018-11-30 | 京东方科技集团股份有限公司 | A kind of method for compressing image, image reconstructing method, apparatus and system |
| CN106297653B (en) * | 2016-10-28 | 2018-08-31 | 重庆工商职业学院 | A kind of LED screen pixel brightness correcting method and its system based on image procossing |
| CN107423709A (en) * | 2017-07-27 | 2017-12-01 | 苏州经贸职业技术学院 | A kind of object detection method for merging visible ray and far infrared |
| CN107423433B (en) * | 2017-08-03 | 2020-07-10 | 聚好看科技股份有限公司 | Data sampling rate control method and device |
| CN107529061B (en) * | 2017-08-06 | 2019-09-06 | 西南交通大学 | Video error coverage method based on compressed sensing and Information hiding |
| CN109903308B (en) * | 2017-12-08 | 2021-02-26 | 百度在线网络技术（北京）有限公司 | Method and device for acquiring information |
| CN108038872B (en) * | 2017-12-22 | 2021-08-31 | 中国海洋大学 | Dynamic and static target detection and real-time compressed sensing tracking research method |
| CN109087267B (en) * | 2018-08-15 | 2021-08-31 | 苏州蛟视智能科技有限公司 | Target detection method based on compressed sensing |
| CN115297288B (en) * | 2022-09-30 | 2023-03-07 | 汉达科技发展集团有限公司 | Monitoring data storage method for driving simulator |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8326240B1 (en) * | 2010-09-27 | 2012-12-04 | Rockwell Collins, Inc. | System for specific emitter identification |
| CN103473792A (en) * | 2013-09-11 | 2013-12-25 | 清华大学深圳研究生院 | Method for detecting moving target |
| CN103810722A (en) * | 2014-02-27 | 2014-05-21 | 云南大学 | Moving target detection method combining improved LBP (Local Binary Pattern) texture and chrominance information |
| CN103903242A (en) * | 2014-04-14 | 2014-07-02 | 苏州经贸职业技术学院 | Self-adaption target compression sensing, fusing and tracking method based on video sensor network |
| CN103996207A (en) * | 2014-04-28 | 2014-08-20 | 清华大学深圳研究生院 | Object tracking method |
| CN104010168A (en) * | 2014-06-13 | 2014-08-27 | 东南大学 | Non-overlapping vision field multi-camera monitoring network topology self-adaptation learning method |
-
2015
- 2015-01-19 CN CN201510023864.XA patent/CN104599290B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8326240B1 (en) * | 2010-09-27 | 2012-12-04 | Rockwell Collins, Inc. | System for specific emitter identification |
| CN103473792A (en) * | 2013-09-11 | 2013-12-25 | 清华大学深圳研究生院 | Method for detecting moving target |
| CN103810722A (en) * | 2014-02-27 | 2014-05-21 | 云南大学 | Moving target detection method combining improved LBP (Local Binary Pattern) texture and chrominance information |
| CN103903242A (en) * | 2014-04-14 | 2014-07-02 | 苏州经贸职业技术学院 | Self-adaption target compression sensing, fusing and tracking method based on video sensor network |
| CN103996207A (en) * | 2014-04-28 | 2014-08-20 | 清华大学深圳研究生院 | Object tracking method |
| CN104010168A (en) * | 2014-06-13 | 2014-08-27 | 东南大学 | Non-overlapping vision field multi-camera monitoring network topology self-adaptation learning method |
Non-Patent Citations (2)
| Title |
|---|
| 应用改进混合高斯模型的运动目标检测;张燕平;《计算机工程与应用》;20101201;第46卷(第34期);第155-158页 * |
| 改进混合高斯模型的运动目标检测算法;华媛蕾 等;《计算机应用》;20140210;第34卷(第2期);第580-584页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104599290A (en) | 2015-05-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104599290B (en) | Video sensing node-oriented target detection method | |
| CN103258332B (en) | A kind of detection method of the moving target of resisting illumination variation | |
| CN107749067A (en) | Fire hazard smoke detecting method based on kinetic characteristic and convolutional neural networks | |
| WO2017000465A1 (en) | Method for real-time selection of key frames when mining wireless distributed video coding | |
| CN103942557B (en) | A kind of underground coal mine image pre-processing method | |
| CN101470809B (en) | Moving object detection method based on expansion mixed gauss model | |
| CN110728183B (en) | Human body action recognition method of neural network based on attention mechanism | |
| CN102184551A (en) | Automatic target tracking method and system by combining multi-characteristic matching and particle filtering | |
| CN108416266A (en) | A kind of video behavior method for quickly identifying extracting moving target using light stream | |
| CN109902565B (en) | Multi-feature fusion human behavior recognition method | |
| CN103971384B (en) | Node cooperation target tracking method of wireless video sensor | |
| CN103150738A (en) | Detection method of moving objects of distributed multisensor | |
| CN105469423B (en) | A kind of online method for tracking target based on continuous attraction sub-neural network | |
| CN116343330A (en) | Abnormal behavior identification method for infrared-visible light image fusion | |
| CN105095898B (en) | A kind of targeted compression cognitive method towards real-time vision system | |
| CN114897728A (en) | Image enhancement method and device, terminal equipment and storage medium | |
| CN110503017A (en) | Wisdom energy conservation occupancy detection system and method based on image procossing | |
| Zhang et al. | A novel framework for background subtraction and foreground detection | |
| CN115113165A (en) | Radar echo extrapolation method, device and system | |
| CN109740527B (en) | Image processing method in video frame | |
| CN110390293A (en) | A kind of Video object segmentation algorithm based on high-order energy constraint | |
| CN109919964A (en) | The method that Gaussian Background modeling technique based on mathematical morphology carries out image procossing | |
| CN104574400A (en) | Remote sensing image segmenting method based on local difference box dimension algorithm | |
| CN106355566A (en) | Smoke and flame detection method applied to fixed camera dynamic video sequence | |
| CN114022371B (en) | Defogging device and defogging method based on space and channel attention residual error network |
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 | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20201028 Address after: 870 Binhe Road, Suzhou hi tech Zone, Suzhou, Jiangsu Province Patentee after: NOVA ELECTRONICS Co.,Ltd. Address before: 215009 No. 287, Xuefu Road, Suzhou, Jiangsu Patentee before: SUZHOU INSTITUTE OF TRADE & COMMERCE |
|
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20220331 Address after: Room 107, No. 56, Jinzhai Road, Yangchenghu Town, Xiangcheng District, Suzhou, Jiangsu 215000 Patentee after: Jiangsu yingshida Electric Technology Co.,Ltd. Address before: 215011 No. 870 Binhe Road, Suzhou high tech Zone, Suzhou, Jiangsu Patentee before: NOVA ELECTRONICS Co.,Ltd. |
|
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20240227 Address after: 472000, No. 6, West Intersection of Jing'er Road and Weiliu Road, Yuwang Road Street, Urban Rural Integration Demonstration Zone, Sanmenxia City, Henan Province Patentee after: Sanmenxia Yunshuo Intelligent Technology Co.,Ltd. Country or region after: China Address before: Room 107, No. 56, Jinzhai Road, Yangchenghu Town, Xiangcheng District, Suzhou, Jiangsu 215000 Patentee before: Jiangsu yingshida Electric Technology Co.,Ltd. Country or region before: China |