CN106997598A - The moving target detecting method merged based on RPCA with three-frame difference - Google Patents
The moving target detecting method merged based on RPCA with three-frame difference Download PDFInfo
- Publication number
- CN106997598A CN106997598A CN201710009254.3A CN201710009254A CN106997598A CN 106997598 A CN106997598 A CN 106997598A CN 201710009254 A CN201710009254 A CN 201710009254A CN 106997598 A CN106997598 A CN 106997598A
- Authority
- CN
- China
- Prior art keywords
- frame
- difference
- rpca
- video
- background
- 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.)
- Pending
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
-
- 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
-
- 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
- G06T2207/20212—Image combination
- G06T2207/20224—Image subtraction
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 moving target detecting method merged based on RPCA with three-frame difference, is comprised the following steps：First, the monitor video of input is read as picture one by one and preserved；Secondly, the video present frame picture of processing needed for reading simultaneously is converted into gray level image；Then, background extracting is carried out by RPCA methods to the video present frame that is read, then, using the background of the video present frame extracted as three-frame difference intermediate frame, the former frame of video as three-frame difference former frame, video present frame be used as a later frame of three-frame difference carry out it is adjacent between difference；Finally, phase "AND" after thresholding is carried out to gained difference result and obtains final detection prospect, threshold value herein is to contrast gained through many experiments；The present invention can extract complete, accurate sport foreground in complex background environment.
Description
Technical field
Melt the present invention relates to machine vision and digital technical field of image processing, more particularly to based on RPCA with three-frame difference
The moving target detecting method of conjunction.
Background technology
The application of intelligent Video Surveillance Technology will greatly promote value of the safety monitoring in terms of social public security is safeguarded.
Human body unusual checking in monitor video is the important component of intelligent monitor system, can be to different in monitor video
Chang Hangwei carries out early warning.Moving object detection in video is the basis of human body unusual checking, therefore, to being moved in video
The research of object detection method has critically important realistic meaning.
Traditional moving object detection algorithm includes frame difference method, optical flow method and background subtraction method.In eighties of last century seventies
End, Jain et al. proposes using the method for inter-frame difference to extract moving target.Frame-to-frame differences method is to the field comprising moving target
Scape has stronger robustness, and arithmetic speed is fast, but this method can not typically detect all pictures of moving object completely
, usually there is " cavity " phenomenon inside the moving object detected in vegetarian refreshments.
Optical flow method is looked for using the correlation between change of the pixel in image sequence in time-domain and consecutive frame
To previous frame with the corresponding relation that exists between present frame, so as to calculate one kind side of the movable information of object between consecutive frame
Method.Optical flow method is not only suitable for static background, is also applied for the situation of cam movement, but the maximum shortcoming of optical flow method is exactly it
It is computationally intensive, it is not suitable for real-time monitoring system.
Background subtraction method is it is necessary to have background image, and background image must be the change with illumination or external environment condition
And real-time update, therefore the key of background subtraction method is background modeling and its renewal.Wren in 1997 et al. is proposed using single
Gauss carries out background modeling method, whether belongs to prospect using threshold decision pixel.But because the complexity of background, single Gauss
Model can not meet requirement.Then, Stauffer in 1999 et al. proposes the mixed Gaussian background modeling method of classics.For
The problem of mixed Gauss model algorithm is computationally intensive, Zivkovic et al. proposes a kind of calculation of Gauss model number adaptively
Method so that efficiency of algorithm, robustness are lifted.2009, Barnich et al. proposed a kind of novel based on pixel
Moving object detection method, and visual background extraction method (ViBe) is named as, the algorithm is directly to each pixel according to certain
Regular random choose a number of pixel value and carry out background modeling, prospect is then carried out to pixel using Euclidean distance
Shortcoming with the classification of background, but ViBe algorithms is easy generation " ghost " and the incomplete problem of moving object detection.
In recent years, realize that moving object detection causes extensive concern with the method for matrix decomposition.This kind of method is recognized
It can be captured by the background parts in observation video by low-rank matrix.What RPCA considered is such a problem, typically
Our data matrix can include structural information, also comprising noise, then this matrix decomposition can be two matrixes by we
It is added, one is low-rank, and another is sparse.Wherein low-rank matrix just represents the background in moving object detection, sparse
Matrix then represents the sport foreground in moving object detection.
The content of the invention
In order to overcome the problem of traditional Three image difference easily a large amount of " cavities " occurs when carrying out moving object detection, it is considered to
It is fairly simple and be easy to realize within hardware to Three image difference, the present invention is intended to provide merged based on RPCA with three-frame difference
Moving target detecting method, inter-frame difference is added to by video background, so as to avoid shadow of the background pixel for foreground detection
Ring, complete, accurate sport foreground can be extracted in complex background environment.
In order to solve the above-mentioned technical problem, the present invention is achieved through the following technical solutions：
The moving target detecting method merged based on RPCA with three-frame difference, is comprised the following steps：
Step1, video data is read in：The monitor video of input is read as picture one by one and preserved；
When handling video, video is read into the picture I for a frame frame first1,I2,…Ik∈Nm*n, wherein m, n is
Per the size of frame picture；
For the detection of sport foreground in kth frame, the frame of kth -1 and kth frame image I are chosenk-1, Ik；
Step2, gray processing processing：The video present frame picture of processing needed for reading simultaneously is converted into gray level image；
To Ik-1, IkGray processing processing is carried out, gray level image is obtained, is designated as I1, I3；
Step3, RPCA background extracting：RPCA methods progress background extracting is passed through to the video present frame read；
By IkBackground recovery is carried out by RPCA methods, so as to obtain representing the low-rank matrix A of backgroundk, it is designated as I2；
Difference between Step4, neighbour：By the background I of the video present frame extracted in Step32As the centre of three-frame difference
Frame, the former frame I of video1It is used as the former frame of three-frame difference, video present frame I3As between a later frame progress neighbour of three-frame difference
Difference；
Solve difference image diff1 and diff2：
Diff1=| I1(x,y)-I2(x,y)|
Diff2=| I3(x,y)-I2(x,y)| (1)
Step5, thresholding：Thresholding processing is carried out to the difference image of gained：
Pass through experiment, it has been found that：When the threshold value taken is more than 10, sport foreground testing result is imperfect；When being taken
When threshold value is less than 10, most of background pixel is sport foreground by flase drop, therefore, and threshold value T is taken as 10 by the present invention；
Step6：Logical "and" operation is carried out to the binary image obtained by difference twice；
It is final to obtain moving object detection prospect f (x, y).
RPCA background extractings are that the low-rank square in RPCA is solved with non-precision augmentation Lagrange multiplier in the Step3
Battle array.
Difference is that consecutive frame picture is subtracted each other between neighbour in the Step4, and the absolute value of gained difference is difference result.
Thresholding processing determines that difference acquired results are converted into bianry image by suitable threshold value T in the Step5.
RPCA methods can accurately recover the background in monitor video, therefore, and the present invention considers the base in Three image difference
On plinth, merge RPCA background extractings to be monitored the moving object detection in video.Pass through the accurate background for extracting RPCA
Former frame of video carry out it is adjacent between difference, it is to avoid traditional Three image difference is for background is more complicated or gray processing rear backdrop picture
Flase drop problem and " cavity " problem that the element situation close with foreground pixel is pre-existed.
Beneficial effects of the present invention：
Method proposed by the invention, by the way that the background image in video is added into inter-frame difference, so as to eliminate background
Influence of the pixel to foreground detection effect, solves " cavity " problem that traditional Three image difference is brought, it also avoid
Traditional moving object detection algorithm flase drop problem that prospect is produced with background pixel when more close in complex scene, proposes calculation
Method accurately can extract moving target in complex background environment, and actual detection is disclosure satisfy that in terms of integrality and accuracy
It is required that.
Brief description of the drawings
The moving target detecting method schematic diagram that Fig. 1 is merged for the present invention based on RPCA with three-frame difference.
Embodiment
With reference to the accompanying drawings and with reference to the Examples detail present invention.
As shown in Figure 1：The moving target detecting method merged based on RPCA with three-frame difference, is comprised the following steps：
Step1, video data is read in：The monitor video of input is read as picture one by one and preserved；
When handling video, video is read into the picture I for a frame frame first1,I2,…Ik∈Nm*n, wherein m, n is
Per the size of frame picture；
For the detection of sport foreground in kth frame, the frame of kth -1 and kth frame image I are chosenk-1, Ik；
Step2, gray processing processing：The video present frame picture of processing needed for reading simultaneously is converted into gray level image；
To Ik-1, IkGray processing processing is carried out, gray level image is obtained, is designated as I1, I3；
Step3, RPCA background extracting：To the video present frame read, background extracting is carried out by RPCA methods；
By IkBackground recovery is carried out by RPCA methods, so as to obtain representing the low-rank matrix A of backgroundk, it is designated as I2；
RPCA core concept be the matrix decomposition that will be interfered into low-rank matrix and sparse matrix i.e. D=A+E, wherein
D is the observing matrix in video sequence, and A is the low-rank matrix for representing background；E is the sparse matrix for representing sport foreground.
Difference between Step4, neighbour：By the background I of the video present frame extracted in Step32As the centre of three-frame difference
Frame, the former frame I of video1It is used as the former frame of three-frame difference, video present frame I3As between a later frame progress neighbour of three-frame difference
Difference；
Solve difference image diff1 and diff2：
Diff1=| I1(x,y)-I2(x,y)|
Diff2=| I3(x,y)-I2(x,y)| (1)
Traditional Three image difference is, by difference between the image progress neighbour comprising prospect and background, not account for complex background
The influence brought for foreground detection；The background extracted by RPCA methods is added between neighbour in difference, so as to avoid by the present invention
The influence that background pixel is brought in differential process；
Step5, thresholding：Thresholding processing is carried out to the difference image of gained：
Pass through experiment, it has been found that：When the threshold value taken is more than 10, sport foreground testing result is imperfect；When being taken
When threshold value is less than 10, most of background pixel is sport foreground by flase drop, therefore, and threshold value T is taken as 10 by the present invention；
Step6：Logical "and" operation is carried out to the binary image obtained by difference twice；
It is final to obtain moving object detection prospect f (x, y).
RPCA methods can accurately recover the background in monitor video, therefore, of the invention on the basis of Three image difference,
RPCA background extractings are merged to be monitored the moving object detection in video.By the way that the RPCA accurate backgrounds extracted are regarded with original
Frequency frame carry out it is adjacent between difference, it is to avoid traditional Three image difference is more complicated or gray processing rear backdrop pixel is with before for background
Flase drop problem and " cavity " problem that the close situation of scene element is pre-existed.
RPCA background extractings are that the low-rank square in RPCA is solved with non-precision augmentation Lagrange multiplier in the Step3
Battle array.
Difference is that consecutive frame picture is subtracted each other between neighbour in the Step4, and the absolute value of gained difference is difference result.
Thresholding processing determines suitable threshold T in the Step5, and difference acquired results are converted into bianry image.
Claims (4)
1. the moving target detecting method merged based on RPCA with three-frame difference, it is characterised in that comprise the following steps：
Step1, video data is read in：The monitor video of input is read as picture one by one and preserved；
When handling video, video is read into the picture I for a frame frame first1,I2,…Ik∈Nm*n, wherein m, n is per frame
The size of picture；
For the detection of sport foreground in kth frame, the frame of kth -1 and kth frame image I are chosenk-1, Ik；
Step2, gray processing processing：The video present frame picture of processing needed for reading simultaneously is converted into gray level image；
To Ik-1, IkGray processing processing is carried out, gray level image is obtained, is designated as I1, I3；
Step3, RPCA background extracting：RPCA methods progress background extracting is passed through to the video present frame read；
By IkBackground recovery is carried out by RPCA methods, so as to obtain representing the low-rank matrix A of backgroundk, it is designated as I2；
Difference between Step4, neighbour：By the background I of the video present frame extracted in Step32As the intermediate frame of three-frame difference, depending on
The former frame I of frequency1It is used as the former frame of three-frame difference, video present frame I3It is used as difference between a later frame progress neighbour of three-frame difference；
Solve difference image diff1 and diff2：
Diff1=| I1(x,y)-I2(x,y)|
Diff2=| I3(x,y)-I2(x,y)| (1)
Step5, thresholding：Thresholding processing is carried out to the difference image of gained：
Pass through experiment, it has been found that：When the threshold value taken is more than 10, sport foreground testing result is imperfect；When taken threshold value
During less than 10, most of background pixel is sport foreground by flase drop, therefore threshold value T is taken as 10 by the present invention；
Step6：Logical "and" operation is carried out to the binary image obtained by difference twice；
It is final to obtain moving object detection prospect f (x, y).
2. the moving target detecting method according to claim 1 merged based on RPCA with three-frame difference, it is characterised in that
RPCA background extractings are that the low-rank matrix in RPCA is solved with non-precision augmentation Lagrange multiplier in the Step3.
3. the moving target detecting method according to claim 1 merged based on RPCA with three-frame difference, it is characterised in that
Difference is that consecutive frame picture is subtracted each other between neighbour in the Step4, and the absolute value of gained difference is difference result.
4. the moving target detecting method according to claim 1 merged based on RPCA with three-frame difference, it is characterised in that
Thresholding processing determines that difference acquired results are converted into bianry image by suitable threshold T in the Step5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710009254.3A CN106997598A (en) | 2017-01-06 | 2017-01-06 | The moving target detecting method merged based on RPCA with three-frame difference |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710009254.3A CN106997598A (en) | 2017-01-06 | 2017-01-06 | The moving target detecting method merged based on RPCA with three-frame difference |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106997598A true CN106997598A (en) | 2017-08-01 |
Family
ID=59431476
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710009254.3A Pending CN106997598A (en) | 2017-01-06 | 2017-01-06 | The moving target detecting method merged based on RPCA with three-frame difference |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106997598A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111583357A (en) * | 2020-05-20 | 2020-08-25 | 重庆工程学院 | Object motion image capturing and synthesizing method based on MATLAB system |
| CN112967321A (en) * | 2021-03-05 | 2021-06-15 | 河北工程大学 | Moving object detection method and device, terminal equipment and storage medium |
| CN114327341A (en) * | 2021-12-31 | 2022-04-12 | 江苏龙冠影视文化科技有限公司 | Remote interactive virtual display system |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102831617A (en) * | 2012-07-17 | 2012-12-19 | 聊城大学 | Method and system for detecting and tracking moving object |
| CN103632373A (en) * | 2013-12-09 | 2014-03-12 | 华东交通大学 | Floc detection method combining three-frame differential higher-order statistics (HOS) with OTSU algorithm |
| CN103679749A (en) * | 2013-11-22 | 2014-03-26 | 北京奇虎科技有限公司 | Moving target tracking based image processing method and device |
| CN103778435A (en) * | 2014-01-16 | 2014-05-07 | 大连理工大学 | Pedestrian fast detection method based on videos |
| CN104504087A (en) * | 2014-12-25 | 2015-04-08 | 中国科学院电子学研究所 | Low-rank decomposition based delicate topic mining method |
| CN106056607A (en) * | 2016-05-30 | 2016-10-26 | 天津城建大学 | Monitoring image background modeling method based on robustness principal component analysis |
-
2017
- 2017-01-06 CN CN201710009254.3A patent/CN106997598A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102831617A (en) * | 2012-07-17 | 2012-12-19 | 聊城大学 | Method and system for detecting and tracking moving object |
| CN103679749A (en) * | 2013-11-22 | 2014-03-26 | 北京奇虎科技有限公司 | Moving target tracking based image processing method and device |
| CN103632373A (en) * | 2013-12-09 | 2014-03-12 | 华东交通大学 | Floc detection method combining three-frame differential higher-order statistics (HOS) with OTSU algorithm |
| CN103778435A (en) * | 2014-01-16 | 2014-05-07 | 大连理工大学 | Pedestrian fast detection method based on videos |
| CN104504087A (en) * | 2014-12-25 | 2015-04-08 | 中国科学院电子学研究所 | Low-rank decomposition based delicate topic mining method |
| CN106056607A (en) * | 2016-05-30 | 2016-10-26 | 天津城建大学 | Monitoring image background modeling method based on robustness principal component analysis |
Non-Patent Citations (2)
| Title |
|---|
| 叶立仁等: ""复杂环境下的遗留物检测算法"", 《计算机工程与科学》 * |
| 邱联奎等: ""基于背景减除与三帧差分相融合的运动检测"", 《合肥工业大学学报（自然科学版）》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111583357A (en) * | 2020-05-20 | 2020-08-25 | 重庆工程学院 | Object motion image capturing and synthesizing method based on MATLAB system |
| CN112967321A (en) * | 2021-03-05 | 2021-06-15 | 河北工程大学 | Moving object detection method and device, terminal equipment and storage medium |
| CN114327341A (en) * | 2021-12-31 | 2022-04-12 | 江苏龙冠影视文化科技有限公司 | Remote interactive virtual display system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110599523A (en) | ViBe ghost suppression method fused with interframe difference method | |
| Liu et al. | Anomaly detection in surveillance video using motion direction statistics | |
| Tu et al. | MSR-CNN: Applying motion salient region based descriptors for action recognition | |
| CN106997598A (en) | The moving target detecting method merged based on RPCA with three-frame difference | |
| TW201032180A (en) | Method and device for keeping image background by multiple gauss models | |
| Yang et al. | Background extraction from video sequences via motion-assisted matrix completion | |
| Parui et al. | An efficient violence detection system from video clips using ConvLSTM and keyframe extraction | |
| Han et al. | Object Detection based on Combination of Visible and Thermal Videos using A Joint Sample Consensus Background Model. | |
| Yang et al. | A hierarchical approach for background modeling and moving objects detection | |
| Lee et al. | Real-time pedestrian and vehicle detection in video using 3d cues | |
| Bhat et al. | DALES: Automated tool for detection, annotation, labelling and segmentation of multiple objects in multi-camera video streams | |
| Huang et al. | A novel algorithm for abandoned object detection | |
| Sun et al. | Moving target detection based on multi-feature adaptive background model | |
| Luo et al. | Adaptive moving vehicle detection algorithm based on hybrid background subtraction and frame difference | |
| Guo et al. | Implementation and analysis of moving objects detection in Video Surveillance | |
| Zhou et al. | A shadow elimination method based on color and texture | |
| Jiang et al. | Background subtraction algorithm based on combination of grabcut and improved ViBe | |
| Cong et al. | Multi-Frames Temporal Abnormal Clues Learning Method for Face Anti-Spoofing | |
| Nalabolu et al. | Object Identification in Digital Videos using Color Histogram Bins and Fuzzy C-Means Clustering Techniques | |
| Peng et al. | An efficient shadow removing algorithm based on projection features | |
| Liu et al. | A Background Modeling Approach Based on Visual Background Extractor | |
| Li et al. | New moving human detecting and tracking algorithm | |
| Chen et al. | The control strategy of moving target detection based on interframe difference algorithm | |
| Hong et al. | A level-set based tracking approach for surveillance video with fusion and occlusion | |
| Chopkar et al. | Real time detection of moving object based on FPGA |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170801 |