CN102129688A - Moving target detection method aiming at complex background - Google Patents
Moving target detection method aiming at complex background Download PDFInfo
- Publication number
- CN102129688A CN102129688A CN 201110044730 CN201110044730A CN102129688A CN 102129688 A CN102129688 A CN 102129688A CN 201110044730 CN201110044730 CN 201110044730 CN 201110044730 A CN201110044730 A CN 201110044730A CN 102129688 A CN102129688 A CN 102129688A
- Authority
- CN
- China
- Prior art keywords
- moving target
- pixel
- background
- sigma
- epsiv
- 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
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention provides a moving target detection method aiming at a complex background, relating to detection aiming at a moving target inside the complex background and solving the problems of poor adaptability and robustness of the traditional detection method under complex conditions. The moving target detection method comprises the following steps of: acquiring an M-frame scene image of a video of a scene to be detected, and carrying out interframe differencing; building a mixture Gaussian model for a difference sequence of each pixel point; setting the false alarm rate of the mixture Gaussian model; calculating a detection threshold of each pixel point; and carrying out binarization processing on each pixel point of each frame of scene images according to the detection threshold to obtain an outline of the moving target. The invention is suitable for detecting the moving target under the complex background.
Description
Technical field
The present invention relates to a kind of object detection method at complex background.
Background technology
Video technique has wide prospect in scientific research and engineering field.In the various researchs and application that video image is handled, the motion target detection technology is the technology of a key.Usually the purpose of moving object detection is for moving target is separated from background, is the binary decision problem of moving target and background.
The top priority that video image is handled is the target that detects motion from video image, by the redundant information on time and the space, background and moving Object Segmentation is come out, and how effectively background and target being cut apart is the emphasis of research at present.
Moving object detection under the complex conditions (or background) is the difficult point that the field was handled and understood to video image always, also becomes the serious hindrance of video image processing system practicality and reliability day by day.Because the occasion of various Video Applications is not quite similar, residing environment of moving target and background are ever-changing, and this adaptability and robustness to moving target detecting method is had higher requirement.But from present condition and technical merit, set up strong interference immunity, can adapt to very difficulty of various occasions moving target detecting method various conditions, sane.The moving object detection of superperformance and the method for tracking have appearred much having under given conditions at present.At the detection method of different application scenarios, its reliability and stability more easily realize, are a full of challenges problem yet how to improve detection method for the adaptability and the robustness of complex conditions (or background).
Summary of the invention
The present invention is in order to solve the bad adaptability of conventional detection under complex conditions, the problem of poor robustness, thereby a kind of moving target detecting method at complex background is provided.
A kind of moving target detecting method at complex background, it is realized by following steps:
Step 1, obtain the video of scene to be measured, obtain M frame scene image according to video content.M is the integer more than or equal to 2;
The difference sequence of each pixel in step 3, the scene image that step 2 is obtained is set up mixed Gauss model;
Step 5, the detection threshold that obtains according to step 4 are done binary conversion treatment to each pixel in the difference sequence frame, thereby are obtained the profile of moving target.
The video P of scene to be measured described in the step 1
X, y(t) expression formula is:
P
x,y(t)=T
x,y(t)+B
x,y(t)+L
x,y(t)
In the formula, x, y are the coordinates of pixel, and t is the time of the video of scene to be measured, T
X, y(t) expression moving target, B
X, y(t) be background, L
X, y(t) intensity of expression illumination.
The concrete implication of the mixed Gauss model described in the step 3 is: one with probability 1-ε from
The stochastic variable that obtains, and with probability ε from
The probability density function of the stochastic variable sum that obtains, the probability density function of mixed Gauss model is defined as:
Wherein,
With
Be respectively the Gauss model of two zero-mean unequal variances, ε ∈ [0,1] is the ratio of two gaussian component;
For every group of difference sequence z[n], n=1,2 ..., N, N are positive integer, and above-mentioned formula is rewritten as:
The method that the false alarm rate of mixed Gauss model is set described in the step 4 is: if with the threshold value of Th as differentiation moving target and background, then the pixel greater than Th is a moving target, and the pixel that is less than or equal to Th is a background; False alarm rate is so:
The described detection threshold employing Q function method that calculates each pixel in every frame scene image according to described false alarm rate of step 4.
Beneficial effect: strong, the strong robustness of the adaptability of detection method of the present invention under complex conditions especially is fit to the moving object detection under brightness variation, shade, leaf swing, the abominable weather environment condition.
Description of drawings
Fig. 1 is a schematic flow sheet of the present invention; Fig. 2 is the Gauss model synoptic diagram of two zero-mean unequal variances; Fig. 3 is the mixed Gauss model synoptic diagram; Fig. 4 is the synoptic diagram of false alarm rate.
Embodiment
Embodiment one, this embodiment is described in conjunction with Fig. 1, a kind of moving target detecting method at complex background, it is realized by following steps:
Step 1, obtain the video of scene to be measured, obtain M frame scene image according to video content, M is the integer more than or equal to 2;
The difference sequence of each pixel in step 3, the every frame scene image that obtains according to step 2 is set up mixed Gauss model;
Step 5, the detection threshold that obtains according to step 4 are done binary conversion treatment to each pixel that step 1 obtains in every frame scene image, thereby are obtained the profile of moving target.
The video P of scene to be measured described in the step 1
X, y(t) expression formula is:
P
x,y(t)=T
x,y(t)+B
x,y(t)+L
x,y(t)
In the formula, x, y are the coordinates of pixel, and t is the time of the video of scene to be measured, T
X, y(t) expression moving target, B
X, y(t) be background, L
X, y(t) intensity of expression illumination.
The concrete implication of the mixed Gauss model described in the step 3 is: one with probability 1-ε from
The stochastic variable that obtains, and with probability ε from
The probability density function of the stochastic variable sum that obtains, the probability density function of mixed Gauss model is defined as:
Wherein,
With
Be respectively the Gauss model of two zero-mean unequal variances, ε ∈ [0,1] is the ratio of two gaussian component;
For every group of difference sequence z[n], n=1,2 ..., N, N are positive integer, and above-mentioned formula is rewritten as:
The method that the false alarm rate of mixed Gauss model is set described in the step 4 is: if with the threshold value of Th as differentiation moving target and background, then the pixel greater than Th is a moving target, and the pixel that is less than or equal to Th is a background; False alarm rate is so:
The described detection threshold employing Q function method that calculates each pixel in every frame scene image according to described false alarm rate of step 4.
Principle: 1, modeling object obtains
The video that order is observed is:
P
x,y(t)=T
x,y(t)+B
x,y(t)+L
x,y(t) (1)
Wherein, x, y are the coordinates of pixel, and t is the time of video sequence, T
X, y(t) target of expression motion, B
X, y(t) be background, L
X, y(t) represent the intensity of illumination.Because the variation of the relative real video signal of variation of illumination is a process that becomes slowly, also even | Δ t| → 0, then the changes delta L of illumination
X, y(t) → 0.When change of background is little, the changes delta B of same background
X, y(t) → 0, then have:
ΔP
x,y(t)=ΔT
x,y(t) (2)
This shows gets the movable information that in fact difference sequence that obtains after the difference has represented moving target to observation sequence.For complex background, when | Δ t| → 0, the changes delta B of background
X, y(t) be not tending towards 0.Therefore must adopt suitable method to remove the influence of background.
For one section video sequence, behind the application inter-frame difference, the pixel of arbitrary position all can produce one group of sequence of differences Δ P
X, y(t), x, y wherein are the coordinates of pixel, and t is the time of video sequence.
Modeling method:
If background and moving target vision signal all are considered as stochastic process, suppose the sequence of differences Δ P of same position pixel so
X, y(t) statistical distribution is f
X, y(z), no matter be not difficult to draw for background or target for inter-frame difference, the average statistical of inter-frame difference is 0.For the target of motion, the deviation of inter-frame difference is greater than the deviation of the inter-frame difference of background certainly.
And actual situation is: Δ P
X, y(t) be one group of data at random, have non-Gaussian.So the application mix Gauss model can characterize Δ P more exactly
X, y(t) distribution.
Mixed Gauss model can be considered as one with probability 1-ε from
The stochastic variable that obtains, and with probability ε from
The probability density function (PDF) of the stochastic variable sum that obtains.The PDF of mixed Gauss model is defined as:
Wherein:
With
Be respectively the Gauss model of two zero-mean unequal variances, ε ∈ [0,1] is the ratio of two gaussian component, for sample z[n], n=1,2 ..., N, (3) formula of rewriting is:
Accompanying drawing 2 and accompanying drawing 3 have been listed the synoptic diagram of mixed Gauss model, and mark 21 is among Fig. 2
Gauss model, mark 22 is
Gauss model.
Asking for of model parameter: the square estimation technique
Suppose three parameters (ε, σ of mixed Gauss model
1, σ
2) be unknown, need be according to the sample z[n of some] estimate.Because PDF is the symmetric function about 0, also is even function, then all odd ordered moment are 0.Can list three independently equations by the even-order square:
These three equations are non-linear, make m
2=E{z
2[n] }, m
4=E{z
4[n] } and m
6=E{z
6[n] } represent 2 rank, 4 rank and the 6 rank squares of sample, and make:
So, directly substitution formula (5) can solve:
Therefore, utilize 2 rank, 4 rank and the 6 rank squares of sample to be not difficult to solve u, corresponding v can utilize 2 rank, 4 rank squares and u to estimate to obtain.U and v are in a single day definite, then
With
Can obtain by finding the solution (7) formula
Mixture ratio is
The statistics of target detects
The principle of Threshold detection: false alarm rate
No matter for background or target, the average statistical of inter-frame difference is 0.The deviation of moving target inter-frame difference is greater than the deviation of the inter-frame difference of background certainly.Using classical statistics and etection theory, as the threshold value of distinguishing moving-target signal and background, so greater than the sample of Th, then is that the possibility of target is big with Th; Less than the sample of Th then is that the possibility of background is big.Because the statistical distribution of moving-target and background all is the symmetrical distribution of 0 average, be the dispersion degree difference, mistake will inevitably appear in detection so.Have two class mistakes: the probability that background is judged to be target mistakenly is called false alarm rate; And being judged to be the probability of background mistakenly, target is called false dismissed rate.False-alarm shows the discrete point-like noise outside the target; False dismissal shows that the cavity appears in the inside of target detection.Therefore this method actual detected is the profile of target.For moving-target detected, false alarm rate was particularly important, and false alarm rate can not be too high, otherwise the outer point that disturbs of target is too many, was difficult to accurately obtain the outline of target.If the statistical distribution of background difference is f
X, y(z), then false alarm rate is
If the statistical distribution of moving-target difference is h
X, y(z), then false dismissed rate is
Accompanying drawing 4 is the synoptic diagram of false alarm rate, and wherein curve 41 is the difference profile h of moving-target
X, y(z); Curve 42 is the difference profile f of background
X, y(z).
The notion of false alarm rate in statistics and the etection theory is applied in the moving object detection, no matter background or moving target, its inter-frame difference average statistical is 0.For the target of motion, the deviation of inter-frame difference is greater than the deviation of the inter-frame difference of background certainly.What depart from 0 big pixel correspondence is the outline and the inner structural edge of moving target, the exactly corresponding needed key feature point of these information points.
Threshold T h asks for: the Q function method
The calculating of the cumulative probability density function of gauss hybrid models can realize by the CDF of single Gauss model.A zero-mean list Gauss PDF satisfies
Work as σ
2=1 o'clock, be standard normal PDF, its cumulative probability density function CDF satisfies:
Adopt the Q function representation of describing right tail probability:
This formula representative surpasses the probability of certain set-point.One of the Q function is approximately:
Therefore, the CDF of mixed Gaussian can be expressed as:
F
x,y(z;ε)=(1-ε)Q
x,y(z/σ
1)+εQ
x,y(z/σ
2) (14)
This formula has explicit Analytical Expression equally, therefore is convenient to obtain corresponding detection threshold according to the statistics etection theory by given false alarm rate.
Foreground segmentation:
Use the threshold value that obtains each pixel is done binary conversion treatment.
In the Threshold detection of moving-target, the control false alarm rate, lower false alarm rate makes the false-alarm point in the background present the discrete point-like form that is similar to noise.Therefore take suitable post-processing technologies such as filtering, be not difficult to determine the outline of moving target, and then realize the foreground segmentation of moving-target.
Obtain after the moving target, further combined with suitable context update technology, just to the parameter real-time update of model, variation that just can implementation model real-time follow-up background.
Claims (5)
1. moving target detecting method at complex background, it is characterized in that: it is realized by following steps:
Step 1, obtain the video of scene to be measured, obtain M frame scene image according to video content, M is the integer more than or equal to 2;
Step 2, the M frame scene image that step 1 is obtained are made inter-frame difference, obtain the difference sequence of each pixel in the scene image;
The difference sequence of each pixel in step 3, the scene image that step 2 is obtained is set up mixed Gauss model;
Step 4, the false alarm rate of requirement is set, and calculates the detection threshold of each pixel in the scene image according to the mixed Gauss model that step 3 obtains;
Step 5, the detection threshold that obtains according to step 4 are done binary conversion treatment to each pixel in the difference sequence frame, thereby are obtained the profile of moving target.
2. a kind of moving target detecting method at complex background according to claim 1 is characterized in that the video P of scene to be measured described in the step 1
X, y(t) expression formula is:
P
x,y(t)=T
x,y(t)+B
x,y(t)+L
x,y(t)
In the formula, x, y are the coordinates of pixel, and t is the time of the video of scene to be measured, T
X, y(t) expression moving target, B
X, y(t) be background, L
X, y(t) intensity of expression illumination.
3. a kind of moving target detecting method at complex background according to claim 1 is characterized in that the concrete implication of the mixed Gauss model described in the step 3 is: one with probability 1-ε from
The stochastic variable that obtains, and with probability ε from
The probability density function of the stochastic variable sum that obtains, the probability density function of mixed Gauss model is defined as:
Wherein,
With
Be respectively the Gauss model of two zero-mean unequal variances, ε ∈ [0,1] is the ratio of two gaussian component;
For (x, y) pixel difference sequence z[n], n=1,2 ..., N, N are positive integer, and above-mentioned formula is rewritten as:
4. a kind of moving target detecting method at complex background according to claim 1 is characterized in that the false alarm rate set described in the step 4 and the pass of detection threshold are:
5. a kind of moving target detecting method at complex background according to claim 1 is characterized in that the described detection threshold employing Q function method that calculates each pixel in every frame scene according to the false alarm rate of setting of step 4.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201110044730A CN102129688B (en) | 2011-02-24 | 2011-02-24 | Moving target detection method aiming at complex background |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201110044730A CN102129688B (en) | 2011-02-24 | 2011-02-24 | Moving target detection method aiming at complex background |
Publications (2)
Publication Number | Publication Date |
---|---|
CN102129688A true CN102129688A (en) | 2011-07-20 |
CN102129688B CN102129688B (en) | 2012-09-05 |
Family
ID=44267763
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201110044730A Expired - Fee Related CN102129688B (en) | 2011-02-24 | 2011-02-24 | Moving target detection method aiming at complex background |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN102129688B (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102867312A (en) * | 2012-08-27 | 2013-01-09 | 苏州两江科技有限公司 | Method for extracting moving object from video image |
CN105099759A (en) * | 2015-06-23 | 2015-11-25 | 上海华为技术有限公司 | Detection method and device |
CN106203429A (en) * | 2016-07-06 | 2016-12-07 | 西北工业大学 | Based on the shelter target detection method under binocular stereo vision complex background |
CN107452005A (en) * | 2017-08-10 | 2017-12-08 | 中国矿业大学(北京) | A kind of moving target detecting method of jointing edge frame difference and gauss hybrid models |
CN111008978A (en) * | 2019-12-06 | 2020-04-14 | 电子科技大学 | Video scene segmentation method based on deep learning |
CN113223043A (en) * | 2021-03-26 | 2021-08-06 | 西安闻泰信息技术有限公司 | Method, device, equipment and medium for detecting moving target |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7336803B2 (en) * | 2002-10-17 | 2008-02-26 | Siemens Corporate Research, Inc. | Method for scene modeling and change detection |
CN101251895A (en) * | 2008-03-13 | 2008-08-27 | 上海交通大学 | Human body tracking method based on gauss mixing model |
CN101470809A (en) * | 2007-12-26 | 2009-07-01 | 中国科学院自动化研究所 | Moving object detection method based on expansion mixed gauss model |
CN101635026A (en) * | 2008-07-23 | 2010-01-27 | 中国科学院自动化研究所 | Method for detecting derelict without tracking process |
CN101686338A (en) * | 2008-09-26 | 2010-03-31 | 索尼株式会社 | System and method for partitioning foreground and background in video |
-
2011
- 2011-02-24 CN CN201110044730A patent/CN102129688B/en not_active Expired - Fee Related
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7336803B2 (en) * | 2002-10-17 | 2008-02-26 | Siemens Corporate Research, Inc. | Method for scene modeling and change detection |
CN101470809A (en) * | 2007-12-26 | 2009-07-01 | 中国科学院自动化研究所 | Moving object detection method based on expansion mixed gauss model |
CN101251895A (en) * | 2008-03-13 | 2008-08-27 | 上海交通大学 | Human body tracking method based on gauss mixing model |
CN101635026A (en) * | 2008-07-23 | 2010-01-27 | 中国科学院自动化研究所 | Method for detecting derelict without tracking process |
CN101686338A (en) * | 2008-09-26 | 2010-03-31 | 索尼株式会社 | System and method for partitioning foreground and background in video |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102867312A (en) * | 2012-08-27 | 2013-01-09 | 苏州两江科技有限公司 | Method for extracting moving object from video image |
CN105099759A (en) * | 2015-06-23 | 2015-11-25 | 上海华为技术有限公司 | Detection method and device |
CN106203429A (en) * | 2016-07-06 | 2016-12-07 | 西北工业大学 | Based on the shelter target detection method under binocular stereo vision complex background |
CN107452005A (en) * | 2017-08-10 | 2017-12-08 | 中国矿业大学(北京) | A kind of moving target detecting method of jointing edge frame difference and gauss hybrid models |
CN111008978A (en) * | 2019-12-06 | 2020-04-14 | 电子科技大学 | Video scene segmentation method based on deep learning |
CN113223043A (en) * | 2021-03-26 | 2021-08-06 | 西安闻泰信息技术有限公司 | Method, device, equipment and medium for detecting moving target |
Also Published As
Publication number | Publication date |
---|---|
CN102129688B (en) | 2012-09-05 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN102129688B (en) | Moving target detection method aiming at complex background | |
CN102881022B (en) | Concealed-target tracking method based on on-line learning | |
CN104657727A (en) | Lane line detection method | |
CN103646257B (en) | A kind of pedestrian detection and method of counting based on video monitoring image | |
WO2015117072A1 (en) | Systems and methods for detecting and tracking objects in a video stream | |
CN103177456A (en) | Method for detecting moving target of video image | |
CN103679704B (en) | Video motion shadow detecting method based on lighting compensation | |
CN106558224B (en) | A kind of traffic intelligent monitoring and managing method based on computer vision | |
CN111402293B (en) | Intelligent traffic-oriented vehicle tracking method and device | |
CN104811586A (en) | Scene change video intelligent analyzing method, device, network camera and monitoring system | |
CN103077520A (en) | Background deduction method for movable vidicon | |
CN104123734A (en) | Visible light and infrared detection result integration based moving target detection method | |
CN102663778B (en) | A kind of method for tracking target based on multi-view point video and system | |
CN105354863A (en) | Adaptive scale image sequence target tracking method based on feature filtering and fast motion detection template prediction | |
CN103559725B (en) | A kind of wireless sensor node optimum choice method of following the tracks of towards vision | |
CN103646242A (en) | Maximally stable extremal region characteristic-based extended target tracking method | |
CN105717491A (en) | Prediction method and prediction device of weather radar echo image | |
CN105447863A (en) | Residue detection method based on improved VIBE | |
Fan et al. | Separation of vehicle detection area using Fourier descriptor under internet of things monitoring | |
CN103400395A (en) | Light stream tracking method based on HAAR feature detection | |
CN104637062A (en) | Target tracking method based on particle filter integrating color and SURF (speeded up robust feature) | |
Tan et al. | A video-based real-time vehicle detection method by classified background learning | |
CN104240268B (en) | A kind of pedestrian tracting method based on manifold learning and rarefaction representation | |
CN103516955A (en) | Invasion detecting method in video monitoring | |
CN104574340A (en) | Video intrusion detection method based on historical images |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
C17 | Cessation of patent right | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20120905 Termination date: 20140224 |