CN107169995A - A kind of adaptive moving target visible detection method - Google Patents
A kind of adaptive moving target visible detection method Download PDFInfo
- Publication number
- CN107169995A CN107169995A CN201710313268.4A CN201710313268A CN107169995A CN 107169995 A CN107169995 A CN 107169995A CN 201710313268 A CN201710313268 A CN 201710313268A CN 107169995 A CN107169995 A CN 107169995A
- Authority
- CN
- China
- Prior art keywords
- mrow
- algorithms
- msub
- algorithm
- pixel
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/254—Analysis of motion involving subtraction of images
-
- 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
- G06T2207/10016—Video; Image sequence
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of adaptive moving target visible detection method, using parallel detection mode, it is detected with multiple moving object detection algorithms for the image sequence of input, real foreground image is estimated according to testing result.Image and the similarity of each testing result are estimated by com-parison and analysis prospect, and then each weight parameter progress when estimating prospect is adaptive, realizes the robust detection of moving target under complex environment.
Description
Technical field
The invention belongs to the technical field that computer vision, video analysis and artificial intelligence are combined, and in particular to one kind is certainly
The moving target visible detection method of adaptation.
Background technology
Traditional moving target visible detection method includes：Optical flow method, frame difference method and background subtraction.Wherein, light stream
Method detection accuracy is high, but it calculates complicated and poor anti jamming capability, is difficult to meet if without specific hardware supported to reality
When the requirement that handles.Frame difference method is most simple efficient a kind of algorithm in these three type algorithms, and it can effectively adapt to environment
, it can be difficult to obtaining complete moving target, easily there is cavitation, Detection results are undesirable during detection in dynamic change.
And background subtraction not only adapts to dynamic environment, and complete target shape is capable of detecting when, it is extensive
Applied in moving object detection neighborhood.The key point of background subtraction is to train an accurately background model, therefore
Different researchers proposes different training methods, obtains different background difference algorithms.Typically there is mixed Gauss model
Algorithm, KDE algorithms, codebook algorithms, ViBe algorithms, SuBSENSE algorithms, AdaDGS algorithms etc..
Although each above-mentioned algorithm can preferably adapt to certain scene, and show good on specific data set,
Due to moving target and the diversity of external environment condition, some specific algorithm is difficult to ensure that at these all to be had under different conditions
There is good detected representation.Although also having other moving object detection algorithms to be suggested in succession, still without a kind of algorithm energy
Different environment are adapted to simultaneously, and obtain good Detection results.
The content of the invention
In order to solve the above-mentioned technical problem, this hair inspects feel method of determining and calculating there is provided a kind of adaptive motion target, realizes not
With the robust detection of moving target under environment.
The technical solution adopted in the present invention is：A kind of adaptive moving target visible detection method, it is characterised in that
Comprise the following steps：
Step 1：Input the first two field picture；
Step 2：Initialize weights omega of the different moving object detection algorithm of N kinds at each pixel xi(x), wherein i
∈ [1, N], x ∈ [1, M], M are the number of image slices vegetarian refreshments；
Step 3：For the different moving object detection algorithm of N kinds, the image sequence of input is detected, obtains different
Foreground image observation { f1,f2,…,fN}；
Step 4：According to ωi(x) with foreground image observation { f1,f2,…,fNThe real foreground image estimate of estimation
Step 5：By foreground image estimateWith { f1,f2,…,fNBe compared, each algorithm is updated in each pixel
Weights omegai(x)；
Step 6：If present image is last frame, terminate algorithm, otherwise input next two field picture, repeat step 3- steps
Rapid 5.
The beneficial effects of the invention are as follows：The present invention is detected to same image sequence simultaneously using many algorithms, and profit
Effective integration is carried out to the result that each algoritic module is detected with the mode of weighted array, real foreground image is estimated.By right
Each algoritic module general performance is estimated and realizes the robust of moving target under complex environment to the adaptive of weight parameter
Detection.In actual applications, compared to other moving object detection algorithms, adaptive motion algorithm of target detection proposed by the present invention
The more complete and less moving target of noise jamming can be obtained.
Brief description of the drawings
Fig. 1 is the general frame figure of adaptive motion algorithm of target detection of the embodiment of the present invention；
Fig. 2 is each parameter physical meaning schematic diagram in index FM calculating process of the embodiment of the present invention；
Fig. 3 is the source images gathered in the embodiment of the present invention；
Fig. 4 is the authentic signature of source images moving target in the embodiment of the present invention；
Fig. 5 is the testing result image of LBAdaSOM algorithms in the embodiment of the present invention；
Fig. 6 is the testing result image of LOBSTER algorithms in the embodiment of the present invention；
Fig. 7 is the testing result image of SuBSENSE algorithms in the embodiment of the present invention；
Fig. 8 is the testing result image of DPZivGMM algorithms in the embodiment of the present invention；
Fig. 9 is the testing result image of ViBe algorithms in the embodiment of the present invention；
Figure 10 is the testing result image of algorithm of the embodiment of the present invention.
Embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, it is right below in conjunction with drawings and Examples
The present invention is further elaborated.It should be appreciated that specific embodiment described herein is only to explain the present invention, not
For limiting the present invention.
A kind of adaptive moving target visible detection method provided see Fig. 1, the present invention, comprises the following steps：
Step 1：Input the first two field picture；
Step 2：Initialize weights omega of the different moving object detection algorithm of N kinds at each pixel xi(x), wherein i
∈ [1, N], x ∈ [1, M], M are the number of image slices vegetarian refreshments；
ωi(x)=1/N；
The present embodiment uses multithreading and distributed variable-frequencypump, and the different moving object detection algorithm of N kinds includes
DPZivGMM algorithms, LBAdaSOM algorithms, codebook algorithms, PAWCS algorithms, FuzzyCho algorithms, FuzzySug algorithms,
MultiCue algorithms, LOBSTER algorithms, ViBe algorithms and SuBSENSE algorithms.
Step 3：For the different moving object detection algorithm of N kinds, the image sequence of input is detected, obtains different
Foreground image observation { f1,f2,…,fN}；
Step 4：According to ωi(x) with foreground image observation { f1,f2,…,fNThe real foreground image estimate of estimation
Step 4.1：Calculate first time foreground image estimate
Wherein fi(x) observation of i-th kind of algorithm at pixel x, f are representedi(x) ∈ { 0,1 }, T1For for the first time
The predefined threshold value of prospect estimation, and T1∈[0,1]；
Step 4.2：By N number of foreground image observation { f1,f2,…,fNAnd first time foreground image estimateCompared
Compared with the general performance FM of each algorithm of calculatingi；
See Fig. 2, wherein P r represent that accuracy rate, Re represent that recall rate, TP (Ture Positive) represent correctly to be examined
The pixel number for prospect, FN (False Negative) is surveyed to represent to be mistakenly detected as pixel number, the FP of background
(False Negative) represents that being mistakenly detected as the pixel number of prospect, TN (True Negative) represents correct
It is detected as the pixel number of background.
Step 4.3：Calculations incorporated algorithm entirety score FMiEach pixel weight γi(x)；
Step 4.4：According to γi(x) second of estimation foreground image estimateMiddle foreground pixel point x is met：
Wherein T2The predefined threshold value estimated for second of prospect, and T2∈[0,1]；
Step 5：By foreground image estimateWith { f1,f2,…,fNBe compared, each algorithm is updated in each pixel
Weights omegai(x)；
1)ωi(x) renewal calculation formula is as follows：
ωi,t(x)=(1- α) ωi,t-1(x)+αMi,t；
Wherein ωi,t(x) it is weight of i-th of the algoritic module of t at pixel x, α is the learning rate of model, α ∈
[0,1], Mi,tFor matching attribute, if fi(x) withMatching, Mi,t=1；Otherwise, Mi,t=0.
Step 6：If present image is last frame, terminate algorithm, otherwise input next two field picture, repeat step 3- steps
Rapid 5.
The present embodiment is based on the platforms of Microsoft Visual Studio 2013, is entered using Opencv computer visions storehouse
Row exploitation.Have 10 kinds for adaptive algorithm, be respectively DPZivGMM algorithms, LBAdaSOM algorithms, codebook algorithms,
PAWCS algorithms, FuzzyCho algorithms, FuzzySug algorithms, MultiCue algorithms, LOBSTER algorithms, ViBe algorithms and
Threshold value T in SuBSENSE algorithms, adaptive process1And T2It is disposed as 0.4.
Fig. 3 is the source images gathered in the embodiment of the present invention, and Fig. 4 is the true of source images moving target in the embodiment of the present invention
Real mark, Fig. 5 is the testing result image of LBAdaSOM algorithms in the embodiment of the present invention, during Fig. 6 is the embodiment of the present invention
The testing result image of LOBSTER algorithms, Fig. 7 is the testing result image of SuBSENSE algorithms in the embodiment of the present invention, and Fig. 8 is
The testing result image of DPZivGMM algorithms in the embodiment of the present invention, Fig. 9 is the detection knot of ViBe algorithms in the embodiment of the present invention
Fruit image, Figure 10 is the testing result image of algorithm of the embodiment of the present invention.
Wherein LBAdaSOM and DPZivGMM algorithms are at detection image sequence 1 and 3, and noise is serious.LOBSTER is in detection
Moving target is lost during sequence 2 and 3, missing inspection is serious.SuBSENSE algorithms equally lost motion mesh in detection sequence 3
Mark, and flase drop is serious in detection sequence 1.ViBe algorithms still have certain missing inspection problem at detection sequence 1 and 2.
And embodiment algorithm is ideal to the testing result of 3 image sequences, can also have while complete moving target is obtained
Effect suppresses noise jamming.
It should be appreciated that the part that this specification is not elaborated belongs to prior art.
It should be appreciated that the above-mentioned description for preferred embodiment is more detailed, therefore it can not be considered to this
The limitation of invention patent protection scope, one of ordinary skill in the art is not departing from power of the present invention under the enlightenment of the present invention
Profit is required under protected ambit, can also be made replacement or be deformed, each fall within protection scope of the present invention, this hair
It is bright scope is claimed to be determined by the appended claims.
Claims (5)
1. a kind of adaptive moving target visible detection method, it is characterised in that comprise the following steps：
Step 1：Input the first two field picture；
Step 2：Initialize weights omega of the different moving object detection algorithm of N kinds at each pixel xi(x), wherein i ∈ [1,
N], x ∈ [1, M], M is the number of image slices vegetarian refreshments；
Step 3：For the different moving object detection algorithm of N kinds, the image sequence of input is detected, before acquisition is different
Scape image observation value { f1,f2,…,fN}；
Step 4：According to ωi(x) with foreground image observation { f1,f2,…,fNThe real foreground image estimate of estimation
Step 5：By foreground image estimateWith { f1,f2,…,fNBe compared, update weight of each algorithm in each pixel
ωi(x)；
Step 6：If present image is last frame, terminates algorithm, otherwise input next two field picture, repeat step 3- steps 5.
2. adaptive moving target visible detection method according to claim 1, it is characterised in that：N described in step 2
Kind different moving object detection algorithms include DPZivGMM algorithms, LBAdaSOM algorithms, codebook algorithms, PAWCS algorithms,
FuzzyCho algorithms, FuzzySug algorithms, MultiCue algorithms, LOBSTER algorithms, ViBe algorithms and SuBSENSE algorithms.
3. adaptive moving target visible detection method according to claim 1, it is characterised in that：Described in step 2
ωi(x)=1/N.
4. adaptive moving target visible detection method according to claim 1, it is characterised in that step 4 it is specific
Realize, including following sub-step：
Step 4.1：Calculate first time foreground image estimate
Wherein fi(x) observation of i-th kind of algorithm at pixel x, f are representedi(x) ∈ { 0,1 }, T1Estimate for first time prospect
Predefined threshold value, and T1∈[0,1]；
Step 4.2：By N number of foreground image observation { f1,f2,…,fNAnd first time foreground image estimateIt is compared, counts
Calculate the general performance FM of each algorithmi；
<mrow>
<msub>
<mi>FM</mi>
<mi>i</mi>
</msub>
<mo>=</mo>
<mn>2</mn>
<mo>&CenterDot;</mo>
<mfrac>
<mrow>
<mi>Pr</mi>
<mo>&CenterDot;</mo>
<mi>Re</mi>
</mrow>
<mrow>
<mi>Pr</mi>
<mo>+</mo>
<mi>Re</mi>
</mrow>
</mfrac>
<mo>=</mo>
<mfrac>
<mrow>
<mn>2</mn>
<mo>&CenterDot;</mo>
<mi>T</mi>
<mi>P</mi>
</mrow>
<mrow>
<mn>2</mn>
<mo>&CenterDot;</mo>
<mi>T</mi>
<mi>P</mi>
<mo>+</mo>
<mi>F</mi>
<mi>P</mi>
<mo>+</mo>
<mi>F</mi>
<mi>N</mi>
</mrow>
</mfrac>
<mo>;</mo>
</mrow>
Wherein Pr represents that accuracy rate, Re represent that recall rate, TP represent to be correctly detected pixel number for prospect, FN and represent quilt
Error detection is the pixel number of background, FP represents that being mistakenly detected as the pixel number of prospect, TN represents correctly to be examined
Survey the pixel number for background；
Step 4.3：Calculations incorporated algorithm entirety score FMiEach pixel weight γi(x)；
<mrow>
<msub>
<mi>&gamma;</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mrow>
<msub>
<mi>FM</mi>
<mi>i</mi>
</msub>
<mo>&CenterDot;</mo>
<msub>
<mi>w</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>N</mi>
</munderover>
<msub>
<mi>FM</mi>
<mi>i</mi>
</msub>
<mo>&CenterDot;</mo>
<msub>
<mi>w</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<mo>;</mo>
</mrow>
Step 4.4：According to γi(x) second of estimation foreground image estimate Middle foreground pixel point x is met：
Wherein T2The predefined threshold value estimated for second of prospect, and T2∈[0,1]。
5. adaptive moving target visible detection method according to claim 4, it is characterised in that in step 5, weight
ωi(x) renewal calculation formula is as follows：
ωi,t(x)=(1- α) ωi,t-1(x)+αMi,t；
Wherein ωi,t(x) it is weight of i-th of the algorithm of t at pixel x, α is learning rate, α ∈ [0,1], Mi,tFor matching
The factor, if fi(x) withMatching, Mi,t=1；Otherwise, Mi,t=0.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710313268.4A CN107169995B (en) | 2017-05-05 | 2017-05-05 | Self-adaptive moving target visual detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710313268.4A CN107169995B (en) | 2017-05-05 | 2017-05-05 | Self-adaptive moving target visual detection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107169995A true CN107169995A (en) | 2017-09-15 |
| CN107169995B CN107169995B (en) | 2020-03-10 |
Family
ID=59812781
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710313268.4A Active CN107169995B (en) | 2017-05-05 | 2017-05-05 | Self-adaptive moving target visual detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107169995B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101673404A (en) * | 2009-10-19 | 2010-03-17 | 北京中星微电子有限公司 | Target detection method and device |
| US20140307056A1 (en) * | 2013-04-15 | 2014-10-16 | Microsoft Corporation | Multimodal Foreground Background Segmentation |
| US9454819B1 (en) * | 2015-06-03 | 2016-09-27 | The United States Of America As Represented By The Secretary Of The Air Force | System and method for static and moving object detection |
-
2017
- 2017-05-05 CN CN201710313268.4A patent/CN107169995B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101673404A (en) * | 2009-10-19 | 2010-03-17 | 北京中星微电子有限公司 | Target detection method and device |
| US20140307056A1 (en) * | 2013-04-15 | 2014-10-16 | Microsoft Corporation | Multimodal Foreground Background Segmentation |
| US9454819B1 (en) * | 2015-06-03 | 2016-09-27 | The United States Of America As Represented By The Secretary Of The Air Force | System and method for static and moving object detection |
Non-Patent Citations (2)
| Title |
|---|
| MASSIMO CAMPLANI ET.AL: "Multi-sensor background subtraction by fusing multiple region-based probabilistic classifiers", 《PATTERN RECOGNITION LETTERS》 * |
| 王洋: "基于DSP的红外小目标检测技术研究", 《万方数据知识服务平台》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107169995B (en) | 2020-03-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109522819B (en) | Fire image identification method based on deep learning | |
| CN109583342B (en) | Human face living body detection method based on transfer learning | |
| CN110929566B (en) | Human face living body detection method based on visible light and near infrared binocular camera | |
| CN104392468B (en) | Based on the moving target detecting method for improving visual background extraction | |
| CN111627044B (en) | Target tracking attack and defense method based on deep network | |
| CN104200236A (en) | Quick target detection method based on DPM (deformable part model) | |
| CN113591968A (en) | Infrared weak and small target detection method based on asymmetric attention feature fusion | |
| CN113850865A (en) | Human body posture positioning method and system based on binocular vision and storage medium | |
| CN113947589A (en) | Missile-borne image deblurring method based on countermeasure generation network | |
| CN111832484A (en) | Loop detection method based on convolution perception hash algorithm | |
| CN105005798B (en) | One kind is based on the similar matched target identification method of structures statistics in part | |
| CN109801232A (en) | A kind of single image to the fog method based on deep learning | |
| CN111310609B (en) | Video target detection method based on time sequence information and local feature similarity | |
| CN113810611B (en) | Data simulation method and device for event camera | |
| CN110287907A (en) | A kind of method for checking object and device | |
| CN108765468A (en) | A kind of method for tracking target and device of feature based fusion | |
| CN107895145A (en) | Method based on convolutional neural networks combination super-Gaussian denoising estimation finger stress | |
| CN105469428B (en) | A kind of detection method of small target based on morphologic filtering and SVD | |
| CN106934770A (en) | A kind of method and apparatus for evaluating haze image defog effect | |
| CN103839280B (en) | A kind of human body attitude tracking of view-based access control model information | |
| CN105469413B (en) | It is a kind of based on normalization ring weighting without refer to smear restoration image synthesis method for evaluating quality | |
| Ding et al. | Machine learning model for feature recognition of sports competition based on improved TLD algorithm | |
| KR101124560B1 (en) | Automatic object processing method in movie and authoring apparatus for object service | |
| CN104851102A (en) | Infrared small target detection method based on human visual system | |
| CN117495718A (en) | Multi-scale self-adaptive remote sensing image defogging method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |