CN104599291B - Infrared motion target detection method based on structural similarity and significance analysis - Google Patents
Infrared motion target detection method based on structural similarity and significance analysis Download PDFInfo
- Publication number
- CN104599291B CN104599291B CN201510030116.4A CN201510030116A CN104599291B CN 104599291 B CN104599291 B CN 104599291B CN 201510030116 A CN201510030116 A CN 201510030116A CN 104599291 B CN104599291 B CN 104599291B
- Authority
- CN
- China
- Prior art keywords
- image
- gauss
- model
- region
- change
- 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.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10048—Infrared image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20076—Probabilistic image processing
Landscapes
- Image Analysis (AREA)
Abstract
A kind of infrared motion target detection method based on structural similarity and significance analysis of image processing field, using GBVS to source images significance analysis, obtains salient region；Source images are divided into change intense regions and change shoulder by improved SSIM again, and using the method optimizing gauss hybrid models of different learning rates；Closed area finally is detected using the gauss hybrid models after optimization, the overlapping region of the closed area and salient region is the final moving target of present frame.The problem of present invention solves mixed Gaussian algorithm edge blurry, with preferable adaptivity and Detection results.
Description
Technical field
It is specifically that one kind is based on structural similarity and conspicuousness the present invention relates to a kind of technology of image processing field
The infrared motion target detection method of analysis.
Background technology
Moving target detection in real time is succeeding target tracking, and the premise of identification, its effect directly affects the steady of follow-up work
Strong property and accuracy.Moving object detection refers to extract the foreground target that there is relative motion with background in the video sequence,
Motion analysis for higher levels such as subsequent target followings is prepared, and is an important research direction of computer vision field,
It is used widely in fields such as intelligent monitor system, man-machine interactive systems.
Moving object detection detects moving target by analyzing the image sequence that imaging sensor is photographed,
Good basis is laid for the behavior understanding of higher.Moving object detection is the emphasis and difficult point of video sequence analysis.In motion
In the studying for a long period of time of target detection, it has been proposed that including many classic algorithms such as calculus of finite differences, Background difference, optical flow method.Wherein：With
Background subtraction is most widely used, and moving target is detected by obtaining background model and comparing frame difference.The essence of background model
Degree determines the validity of background subtraction, if background modeling process occurrence scene changes, and situations such as imaging device trembles can be serious
The contrast and signal to noise ratio of image are reduced, the identification to infrared target is influenceed.
Background modeling method based on gauss hybrid models, by continuous Gaussian component modeling background information, is recycled
Background information Difference test goes out moving target, can preferably solve multi-modal background problems, is especially suitable for outdoor light and weather
The small and fireballing moving object detection of change.But when gauss hybrid models initialization, new model set up and learning rate not
Timing can all produce diplopia phenomenon.
By the retrieval discovery to prior art, Chinese patent literature CN103810703A discloses (bulletin) day
2014.05.21, a kind of tunnel video moving object detection method based on image procossing is disclosed, this method includes following step
Suddenly：Set up initial back-ground model；Set up dynamic in real time and update background model；Construct partial structurtes similarity measure function；Construction
Local gray level statistical measurement function；Motion is extracted according to partial structurtes similarity measure function and local gray-scale statistical measure function
Target area.But the technology merely determines whether moving target, it is discrete to extract moving target by algorithm, not comprising original
Beginning clarification of objective information, it is impossible to provide support for the follow-up further processing to moving target.
The content of the invention
The present invention can not detect the change of background in time for traditional Gauss mixed model, by its detection obtain it is red
Outer target includes false profile, is difficult the deficiency accurately identified, proposes a kind of red based on structural similarity and significance analysis
Outer moving target detecting method and system, the spatial information for combining sign target area discrete pixels point by watershed algorithm are obtained
To enclosed region, then pass through point based on PCNN (Pulse Coupled Neural Network, Pulse Coupled Neural Network)
Cut algorithm eliminate diplopia, finally detect complete infrared motion target, so as to preferably extract foreground target, with compared with
Good effect and stronger robustness.
The present invention is achieved by the following technical solutions：
It is sharp first the present invention relates to a kind of infrared motion target detection method based on structural similarity and significance analysis
It is notable to source images with GBVS (Graph-Based Visual Saliency, the saliency parser based on graph theory)
Property analysis, obtain salient region；By source images, by improved SSIM, (Structural similarity, structure is similar again
Degree algorithm) change intense regions and change shoulder are divided into, and using the method optimizing Gaussian Mixture mould of different learning rates
Type；Closed area finally is detected using the gauss hybrid models after optimization, the overlay region of the closed area and salient region
Domain is the final moving target of present frame.
Described salient region, carries out significance analysis to infrared sequence image using GBVS and obtains.
Described change intense regions and change shoulder, are obtained in the following manner：
1) structure for asking for the local block of adjacent two frames figure in infrared sequence image using improved SSIM algorithms is similar
Degree；
2) the structural similarity statistical value of local block is calculated using sliding window mode, and generation is based on this basis
The statistical chart of infrared sequence image；
3) using image local situation of change in CDF mode counting statistics figures, smooth rear distribution curve maximum curvature is found
The corresponding cumulative number Au of pointmax, Au will be more than in statistical chartmaxPoint constitute region divide into change intense regions, remaining
For change shoulder.
Described optimization gauss mixed model refers to：For change intense regions using larger learning rate, change flat zone
Domain uses smaller learning rate.
The present invention relates to a kind of system for realizing the above method, including：Salient region extraction module, structural similarity point
Generic module, Gauss model update module and closed area detection module, wherein：Structural similarity sort module is divided the image into
Background area and target area, Gauss model update module are connected with structural similarity sort module, and different study are respectively adopted
Rate updates background area and target area, and transmitting discrete moving target information, closed area detection module updates with Gauss model
Model, which is connected, transmits closure target information, and salient region extraction module is connected to transmit with closed area detection module accurately to be transported
Moving-target information.
Technique effect
Compared with prior art, the present invention, to solve the replacement problem of background model, utilizes structure in terms of model learning
Similarity algorithm by infrared image background be divided into graded substantially with slow region, different learning rates are set respectively and are updated
Gauss hybrid models, it is ensured that the Stability and veracity of model, with the position in quick obtaining infrared motion target region；In target
Context of detection, the enclosed region of target is obtained using the watershed algorithm based on spatial information, recycles the vision based on figure
Conspicuousness (Graph-Based Visual Salience, GBVS) algorithm removes diplopia, finally gives complete moving target.
Test result indicates that, the problem of this method solves mixed Gaussian algorithm edge blurry well, with preferable adaptivity
And Detection results.
Brief description of the drawings
Fig. 1 is cumulative distribution function figure；
Fig. 2 is schematic flow sheet of the present invention；
Fig. 3 is the effect diagram of embodiment 1；
In figure：(a), (b) is respectively the frame of infrared sequence image the 18th, 19, and (c), (d) is respectively constant learning rate and becomes
The target area of habit rate detection, (e)-(h) is respectively by frame difference method, background subtraction and the present invention between watershed algorithm, neighbour
The testing result that method is obtained.
Fig. 4 is to split obtained accurate target region by hand by Fig. 3 (b).
Embodiment
Embodiments of the invention are elaborated below, the present embodiment is carried out lower premised on technical solution of the present invention
Implement, give detailed embodiment and specific operating process, but protection scope of the present invention is not limited to following implementations
Example.
Embodiment 1
As shown in Fig. 2 the present embodiment is A to infrared sequence imaget(t=1 ... n) carries out following handle：
The first step：Using GBVS algorithms to infrared sequence image AtSignificance analysis is carried out, salient region B is obtainedt, tool
Body step includes：
1.1) defined nucleotide sequence image AtTwo pixel m of characteristic pattern1(i1,j1) and m2(i2,j2) otherness：
Wherein：Mf() is characterized the pixel in figure.
1.2) obtain constructing each summit in the figure connected entirely, figure after the otherness of any two points in image and represent one
Pixel, each edge represents the weight between two pixels, i.e.,
Wherein：σ is scale factor, m1(i1,j1) and m2(i2,j2) it is two pixels；
1.3) salient region is obtained after all weights are normalized.
Second step：The structure of the local block of adjacent two frames figure in infrared sequence image is asked for using improved SSIM algorithms
Similarity, specific steps include：
2.1) entire image is pressed from top to bottom, if being sequentially divided into from left to right by the sliding window of 7 × 7 pixel sizes
Dry weight folds sub-block, and then carrying out mirror-extended to image when sliding window exceeds image range obtains 7 × 7 sub-blocks, if the chi of image
It is very little be m × n when, then the sub-block number obtained be m × n；
2.2) the sub-block Block (x, y) of same position structural similarity is in image F and image L：
Wherein：(x, y) represents sub-block center pixel, uF、uL、σF、σLImage F and image L sub-block Block is represented respectively
The average and variance of (x, y)；C1、C2Respectively tend to 0 normal amount.
3rd step：The structural similarity statistical value A of local block is calculated using sliding window modet' (x, y), and herein
On the basis of generate the statistical chart based on infrared sequence image：
Dt(x, y)=At'(x,y)-At(x, y),
Wherein：Ones (x, y) is all 1's matrix of centered on (x, y) 7 × 7 sizes, when the structural similarity of local block
During higher than 0.9, it is believed that its intensity of variation is little, At(x, y) is constant, when the structural similarity of sub-block is less than 0.9, with (x, y)
Centered on 7 × 7 statistical value of matrix add 1.
Described statistical chart DtThe gray value bigger region representation regional change degree is more violent in (x, y).
4th step：Utilize CDF (s) (Cumulative Distribution Function, cumulative distribution function) mode
Counting statistics figure DtImage local situation of change in (x, y), the point for finding smooth rear distribution curve maximum curvature is corresponding cumulative
Number Aumax, by statistical chart DtIt is more than Au in (x, y)maxThe region that constitutes of point divide into change intense regions Atr, remaining is change
Shoulder Atg。
Described cumulative distribution function is the function of the localized variation degree of image, that is, seeks image DtPixel value s in (x, y)
The probability P (s) of appearance, cumulative distribution function CDF (s)=P (S≤s) obtains smoothed curve function by 10 difference fittings, such as
Shown in Fig. 1；Obtain D corresponding to the maximum point of the curvature of curvetThe value Au of (x, y)max, by DtPixel value is more than Au in (x, y)max
Point composition region be change intense regions.
5th step：By infrared sequence image At(t=1 ... n) updates Gauss by way of different zones correspondence learning rate
Mixed model, i.e. gauss hybrid models are carried out more in learning process according to the image information of a new frame to weight, average, variance
Newly, and change intense regions are updated with 10 times of learning rates, change shoulder keeps original learning rate α, specific steps
Including：
5.1) pixel for representing moving target is calculated：
Wherein：P(St) it is StThe distribution function of (x, y), St(x, y) represents the pixel value of t width images, ωi,t(x, y) table
Show that t width image slices vegetarian refreshments belongs to the weights of i-th of Gaussian Profile, ui,tFor Gauss model average, Σi,tFor the association of Gauss model
Variance, σi,tFor standard deviation, I is unit matrix, and n is XtDimension, η () be Gauss model probability density function.
5.2) following update is carried out to Gauss model：
Wherein：α is learning rate, represents that moving target incorporates the speed of background, by pixel value St(x, y) and K Gaussian mode
Type is matched, and works as St(x, y) and i-th of Gauss model difference are less than 2.5 times of standard deviation sigmasi,tWhen think to match with "current" model,
That is MtMatching is represented when=1, the M when mismatchingt=0, and accordingly reduce weight and average and variance are not updated.
5.3) due to StThe Gauss model that (x, y) is matched the most has maximum weights and minimum standard deviation, by K
Gauss model is according to ωi,t/σi,tThe arrangement of ratio descending, the Gauss model at the top of sequence most possibly describes the stable back of the body
Scape, and the Gauss model in sequence bottom describes moving target, therefore：
Using preceding Num in K Gauss model as background model, τ represents background Gauss model institute in probability distribution
The number of the minimum scale accounted for, i.e. background model
Work as St(x, y) is all mismatched with K Gauss model, then is x last Gauss model is come using averaget, standard
Poor σi,tWith weights ωi,tThe Gauss model for being respectively set to initial value is replaced；After the completion of updating every time, to weights ωi,tReturned
One change is handled.
6th step：Using the gray value of each pixel as level height value, P (S are calculatedt) spatial frequency SFtObtain discrete
The edge of target, then closed area E is obtained using watershed algorithm connection edget。
Described spatial frequency refers to：The index of grey scale change severe degree in phenogram picture, i.e., gray scale is in plane space
On gradient.
Described watershed algorithm is realized by following iterative manner：
Wherein：X (h) be when level value is h each
Gray value is less than the set of h point in the joint of regional ensemble, i.e. image, and I (p) represents the gray value of image, hminAnd hmaxTable
Diagram is as minimum and highest gray value；T(hmin) corresponding connected domain when gray value is minimum is represented, MIN (h+1) is gray value in h
The joint of all Minimum Areas, Z when+1T(h+1)(X (h)) represents geodetic influence areas of the region X (h) in connected domain T (h+1).
Described closed areaWshedmaxFor area in image most
Big enclosed region, Wshed (f)=D-X (hmax) i.e. in image DtX (h in (x, y)max) supplementary set.
Described geodetic influence area refers to：WhenIt is to be divided into the k region B being connectedi, i=1 ..., k,
Then in A, subset BiGeodetic influence area be defined as：P ∈ A |=dA(p,Bi) ＜ dA(p,B\Bi), geodetic influence area claims
For reception basin, and the border of reception basin then forms watershed.
7th step：By the gauss hybrid models after being updated in the 5th step to the closed area E in the 6th steptDetected,
Gained region and the salient region B obtained in the first steptLap region be present frame final moving target.
The target area that this method is obtained than the result complete and accurate that other method is obtained, as shown in figure 3, by with Fig. 4
Shown in shown standard picture is compared as follows, wherein MI represents coefficient correlation, and QAB/F represents edge gradient information, is worth bigger table
Show obtained moving target and former target closer to.
| MI | QAB/F | |
| This method | 0.2577 | 0.5922 |
| Background subtraction | 0.2459 | 0.4003 |
| Calculus of finite differences between neighbour | 0.0854 | 0.1374 |
Claims (7)
1. a kind of infrared motion target detection method based on structural similarity and significance analysis, it is characterised in that sharp first
With GBVS to source images significance analysis, salient region is obtained；Source images are divided into change by improved SSIM again acute
Strong region and change shoulder, and using the method optimizing gauss hybrid models of different learning rates；Finally utilize after optimization
Gauss hybrid models detect closed area, and the overlapping region of the closed area and salient region is the final fortune of present frame
Moving-target；
Described change intense regions and change shoulder, are obtained in the following manner：
1) structural similarity of the local block of adjacent two frames figure in infrared sequence image is asked for using improved SSIM algorithms；
2) the structural similarity statistical value A of local block is calculated using sliding window modet' (x, y), and generate on this basis
Statistical chart D based on infrared sequence imaget(x,y)；
3) CDF mode counting statistics figures D is utilizedtImage local situation of change in (x, y), finds smooth rear distribution curve curvature most
The corresponding cumulative number Au of big pointmax, by statistical chart DtIt is more than Au in (x, y)maxTo divide into change violent in the point region that constitutes
Region Atr, remaining is change shoulder Atg；
The structural similarity of described local block, is obtained in the following manner：
I) entire image is pressed from top to bottom by the sliding window of 7 × 7 pixel sizes, be sequentially divided into from left to right some overlapping
Sub-block, sliding window exceed image range when then to image carry out mirror-extended obtain 7 × 7 sub-blocks, if the size of image be m ×
During n, then the sub-block number obtained is m × n；
Ii) the sub-block Block (x, y) of same position structural similarity is in image F and image L：
Wherein：(x, y) represents sub-block center pixel, uF, uL,
σF, σLImage F and image L sub-block Block (x, y) average and variance is represented respectively；C1, C2Respectively tend to 0 it is normal
Amount；
The described statistical chart based on infrared sequence image, the structure phase of local block is calculated using sliding window mode
Like degree statistical value At' (x, y), and D is generated on this basist(x, y)=At'(x,y)-At(x, y),Wherein：Ones (x, y) is is with (x, y)
1 matrix of the size of center 7 × 7, when the structural similarity of local block is higher than 0.9, it is believed that its intensity of variation is little, At(x,
Y) constant, when the structural similarity of sub-block is less than 0.9, the statistical value of centered on (x, y) 7 × 7 matrix adds 1.
2. according to the method described in claim 1, it is characterized in that, described salient region, using GBVS algorithms to infrared sequence
Row image AtCarry out significance analysis to obtain, specific steps include：
1.1) defined nucleotide sequence image AtTwo pixel m of characteristic pattern1(i1,j1) and m2(i2,j2) otherness：Wherein：Mf() is characterized the pixel in figure；
1.2) obtain constructing each summit in the figure connected entirely, figure after the otherness of any two points in image and represent a picture
Element, each edge represents the weight between two pixels, i.e.,
Wherein：σ is scale factor, m1(i1,j1) and m2(i2,j2) it is two pixels；
1.3) salient region is obtained after all weights are normalized.
3. according to the method described in claim 1, it is characterized in that, described image local situation of change, by calculating iterated integral
The function of the localized variation degree of cloth function, i.e. image is obtained：Counting statistics figure DtThe probability P that pixel value occurs in (x, y)
(s), cumulative distribution function CDF (s)=P (S≤s) obtains smoothed curve function by 10 difference fittings, and the curvature of curve is most
D corresponding to big pointtThe value Au of (x, y)max, by DtPixel value is more than Au in (x, y)maxThe region of point composition be that change is acute
Strong region.
4. method according to claim 2, it is characterized in that, described optimization gauss mixed model refers to：By infrared sequence
Image updates gauss hybrid models, i.e. the gauss hybrid models root in learning process by way of different zones correspondence learning rate
Weight, average, variance are updated according to the image information of a new frame, and change intense regions are carried out more with 10 times of learning rates
Newly, change shoulder keeps original learning rate α.
5. the method according to claim 2 or 4, it is characterized in that, described optimization gauss mixed model is specifically included：
I) pixel for representing moving target is calculated：Wherein：P
(St) it is StThe distribution function of (x, y), St(x, y) represents the pixel value of t width images, ωi,t(x, y) represents t width image slices
Vegetarian refreshments belongs to the weights of i-th of Gaussian Profile, ui,tFor Gauss model average, ∑i,tFor the covariance of Gauss model, σi,tFor mark
Accurate poor, I is unit matrix, and n is XtDimension, η () be Gauss model probability density function；
Ii following update) is carried out to Gauss model：Wherein：α is learning rate, table
Show that moving target incorporates the speed of background, by pixel value St(x, y) is matched with K Gauss model, and works as St(x, y) is high with i-th
The difference of this model is less than 2.5 times of standard deviation sigmasi,tWhen think to match with "current" model, i.e. MtMatching is represented when=1, when mismatching
Mt=0, and accordingly reduce weight and average and variance are not updated；
Iii) due to StThe Gauss model that (x, y) is matched the most has maximum weights and minimum standard deviation, by K Gauss
Model is according to ωi,t/σi,tThe arrangement of ratio descending, the Gauss model at the top of sequence most possibly describes Steady Background Light, and
Gauss model in sequence bottom describes moving target, and using preceding Num in K Gauss model as background model, τ is represented
The number of background Gauss model minimum scale, i.e. background model shared in probability distribution
Iv S) is worked ast(x, y) is all mismatched with K Gauss model, then is x last Gauss model is come using averaget, standard
Poor σi,tWith weights ωi,tThe Gauss model for being respectively set to initial value is replaced；After the completion of updating every time, to weights ωi,tReturned
One change is handled；
Iv) using the gray value of each pixel as level height value, P (S are calculatedt) spatial frequency SFtObtain the side of dispersive target
Edge, then closed area E is obtained using watershed algorithm connection edget。
6. method according to claim 5, it is characterized in that, described closed areaWshedmaxFor the enclosed region that area in image is maximum, i.e. watershed
Region, Wshed (f)=Dt(x,y)-X(hmax), wherein：Image DtIn (x, y), X (hmax) it is that level value is hmaxWhen regional
Gray value is less than h in union of sets, i.e. imagemaxPoint set；
Described watershed is formed by the border of geodetic influence area, is iterated to calculate and is obtained especially by watershed algorithm：
Wherein：X (h) is regional when level value is h
Gray value is less than the set of h point in union of sets, i.e. image, and I (p) represents the gray value of image, hminAnd hmaxRepresent figure
As minimum and highest gray value；T(hmin) corresponding connected domain when gray value is minimum is represented, MIN (h+1) is gray value in h+1
The joint of all Minimum Areas, ZT(h+1)(X (h)) represents geodetic influence areas of the region X (h) in connected domain T (h+1).
7. a kind of system for realizing any of the above-described claim methods described, it is characterised in that including：Salient region extracts mould
Block, structural similarity sort module, Gauss model update module and closed area detection module, wherein：Structural similarity point
Generic module divides the image into background area and target area, and Gauss model update module is connected with structural similarity sort module,
Different learning rates are respectively adopted and update background area and target area, transmitting discrete moving target information, closed area detection mould
Block is connected with Gauss model more new model transmits closure target information, salient region extraction module and closed area detection module
It is connected to transmit accurately moving target information.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510030116.4A CN104599291B (en) | 2015-01-21 | 2015-01-21 | Infrared motion target detection method based on structural similarity and significance analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510030116.4A CN104599291B (en) | 2015-01-21 | 2015-01-21 | Infrared motion target detection method based on structural similarity and significance analysis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104599291A CN104599291A (en) | 2015-05-06 |
| CN104599291B true CN104599291B (en) | 2017-07-28 |
Family
ID=53125041
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510030116.4A Expired - Fee Related CN104599291B (en) | 2015-01-21 | 2015-01-21 | Infrared motion target detection method based on structural similarity and significance analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104599291B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106251355B (en) * | 2016-08-03 | 2018-12-14 | 江苏大学 | A kind of detection method merging visible images and corresponding night vision infrared image |
| CN108256519B (en) * | 2017-12-13 | 2022-06-17 | 苏州长风航空电子有限公司 | Infrared image significance detection method based on global and local interaction |
| CN109359518A (en) * | 2018-09-03 | 2019-02-19 | 惠州学院 | A kind of moving object recognition methods, system and the warning device of infrared video |
| CN110944199A (en) * | 2019-11-28 | 2020-03-31 | 华侨大学 | Screen content video code rate control method based on space-time perception characteristics |
| CN112927238B (en) * | 2019-12-06 | 2022-07-01 | 四川大学 | Core sequence image annotation method combining optical flow and watershed segmentation |
| CN116092018B (en) * | 2023-04-10 | 2023-08-25 | 同方德诚(山东)科技股份公司 | Fire-fighting hidden danger monitoring method and system based on intelligent building |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102509303A (en) * | 2011-11-22 | 2012-06-20 | 鲁东大学 | Binarization image registration method based on improved structural similarity |
| CN102968782A (en) * | 2012-09-12 | 2013-03-13 | 苏州大学 | Automatic digging method for remarkable objects of color images |
| CN102968786A (en) * | 2012-10-23 | 2013-03-13 | 西北工业大学 | Unsupervised potential target region detection method for remote sensing images |
| CN103077520A (en) * | 2013-01-04 | 2013-05-01 | 浙江大学 | Background deduction method for movable vidicon |
| CN103336947A (en) * | 2013-06-21 | 2013-10-02 | 上海交通大学 | Method for identifying infrared movement small target based on significance and structure |
| JP2014137688A (en) * | 2013-01-16 | 2014-07-28 | Toyota Motor Corp | Object detection apparatus and object detection method, alarm device and alarm method, and driving assistance apparatus and driving assistance method |
-
2015
- 2015-01-21 CN CN201510030116.4A patent/CN104599291B/en not_active Expired - Fee Related
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102509303A (en) * | 2011-11-22 | 2012-06-20 | 鲁东大学 | Binarization image registration method based on improved structural similarity |
| CN102968782A (en) * | 2012-09-12 | 2013-03-13 | 苏州大学 | Automatic digging method for remarkable objects of color images |
| CN102968786A (en) * | 2012-10-23 | 2013-03-13 | 西北工业大学 | Unsupervised potential target region detection method for remote sensing images |
| CN103077520A (en) * | 2013-01-04 | 2013-05-01 | 浙江大学 | Background deduction method for movable vidicon |
| JP2014137688A (en) * | 2013-01-16 | 2014-07-28 | Toyota Motor Corp | Object detection apparatus and object detection method, alarm device and alarm method, and driving assistance apparatus and driving assistance method |
| CN103336947A (en) * | 2013-06-21 | 2013-10-02 | 上海交通大学 | Method for identifying infrared movement small target based on significance and structure |
Non-Patent Citations (4)
| Title |
|---|
| 一种用于CTP版面瑕疵的自适应检测算法;沈凌云等;《长春理工大学学报（自然科学版）》;20141231;第37卷(第6期);第96页左栏第20-36行 * |
| 基于改进分水岭算法的红外图像分割;孙伟等;《红外与激光工程》;20061031;第35卷;第31-37页 * |
| 基于自适应混合高斯模型的时空背景建模;王永忠;《自动化学报》;20090430;第35卷(第4期);第371-378页 * |
| 田洪金等.基于自适应分块和SSIM的运动目标检测.《计算机科学》.2014,第41卷(第2期), * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104599291A (en) | 2015-05-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | Automatic laser profile recognition and fast tracking for structured light measurement using deep learning and template matching | |
| CN104599291B (en) | Infrared motion target detection method based on structural similarity and significance analysis | |
| CN102542289B (en) | Pedestrian volume statistical method based on plurality of Gaussian counting models | |
| CN110033473B (en) | Moving target tracking method based on template matching and depth classification network | |
| CN102214291B (en) | Method for quickly and accurately detecting and tracking human face based on video sequence | |
| CN103971386B (en) | A kind of foreground detection method under dynamic background scene | |
| CN107169985A (en) | A kind of moving target detecting method based on symmetrical inter-frame difference and context update | |
| CN104978567B (en) | Vehicle checking method based on scene classification | |
| Tan et al. | Vehicle detection in high resolution satellite remote sensing images based on deep learning | |
| CN105260749B (en) | Real-time target detection method based on direction gradient binary pattern and soft cascade SVM | |
| CN105528794A (en) | Moving object detection method based on Gaussian mixture model and superpixel segmentation | |
| CN105513053B (en) | One kind is used for background modeling method in video analysis | |
| CN104601964A (en) | Non-overlap vision field trans-camera indoor pedestrian target tracking method and non-overlap vision field trans-camera indoor pedestrian target tracking system | |
| CN102609724B (en) | Method for prompting ambient environment information by using two cameras | |
| CN102147861A (en) | Moving target detection method for carrying out Bayes judgment based on color-texture dual characteristic vectors | |
| CN109919053A (en) | A kind of deep learning vehicle parking detection method based on monitor video | |
| CN106204594A (en) | A kind of direction detection method of dispersivity moving object based on video image | |
| CN104517095A (en) | Head division method based on depth image | |
| CN108074234A (en) | A kind of large space flame detecting method based on target following and multiple features fusion | |
| CN103955949A (en) | Moving target detection method based on Mean-shift algorithm | |
| CN107895379A (en) | The innovatory algorithm of foreground extraction in a kind of video monitoring | |
| CN109377511B (en) | Moving target tracking method based on sample combination and depth detection network | |
| CN106570490A (en) | Pedestrian real-time tracking method based on fast clustering | |
| CN103456009B (en) | Object detection method and device, supervisory system | |
| CN113592911A (en) | Apparent enhanced depth target tracking method |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170728 Termination date: 20180121 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |