CN102243765A - Multi-camera-based multi-objective positioning tracking method and system - Google Patents

Abstract

The invention discloses a multi-camera-based multi-objective positioning tracking method. The method is characterized by comprising the following steps: installing a plurality of cameras at a plurality of visual angles firstly, planning a public surveillance area for the cameras, and calibrating a plurality of height levels; sequentially implementing the steps of foreground extraction, homography matrix calculation, foreground likelihood fusion and multi-level fusion; extracting positioning information which is based on selected a plurality of height levels and obtained in the step of foreground likelihood fusion; processing the positioning information of each level by using the shortest path algorithm so as to obtain the tracking paths of the levels; and after combining with the processing results of foreground extraction, completing the multi-objective three-dimensional tracking. By using the method disclosed by the invention, in the process of tracking, the vanishing points of the plurality of cameras are not required to be calculated, and a codebook model is introduced for the first time for solving the multi-objective tracking problem, thereby improving the accuracy of tracking; and the method has the characteristics of good stability, good instantaneity and high precision.

Description

Multiple goal positioning and tracing method and system based on polyphaser

Technical field

The present invention relates to a kind of multiple goal positioning and tracing method and system based on polyphaser.

Background technology

Multiple target tracking is a hot research problem of computer vision field, is prerequisite and basis that the moving target behavior is understood, all has a wide range of applications in fields such as robot navigation, video understanding, intelligent monitorings.Multiple target tracking is in the specific implementation process, and the most unmanageable is exactly that multiple goal is frequently blocked each other.In this case, a foreground area may belong to a plurality of targets.According to traditional color distribution, information such as shape, we may realize moving target is monitored accurately and followed the tracks of hardly.In recent years, a lot of scholars have done the research of this respect, and the method that is proposed can roughly be divided into two classes: one camera is followed the tracks of and the polyphaser tracking.

Though the tracking based on one camera deals with fairly simple, but because the target in the 3d space has been lost a lot of information in the 2D imaging process, only go monitoring and tracking target to be difficult to well solve occlusion issue, and then can't obtain tracking effect accurately from a visual angle.Therefore, the polyphaser tracking more and more is subjected to researchist's attention, the process that polyphaser is followed the tracks of is to utilize a plurality of cameras from different visual angles a plurality of targets independently to be monitored and followed the tracks of, and the information at each visual angle of effective fusion solves the problem of blocking then, reaches the purpose of tenacious tracking.

Based on the multi-object tracking method of moving target monitoring, the method that is proposed belongs to the one camera tracking in the publication number CN101887587 video monitoring, is taking place under the situation of blocking fully, follows the tracks of stable inadequately.This method moving target monitored results of being based on one camera fully realizes following the tracks of in addition, and the result of moving target monitoring under many circumstances can be inaccurate, and that will inevitably cause the inaccurate of tracking.The method that publication number CN101154289A follows the tracks of based on the 3 d human motion of many orders camera, the motion tracking in the human body three-dimensional skeleton joint that this patent is mainly set forth, the problem that solves with our method is different.

People such as Khan S. on IEEE Transactions on Pattern Analysis and Machine Intelligence, published thesis in 2009 " Tracking Multiple Occluding People by Localizing on Multiple Scene Planes " a kind of polyphaser tracking that does not need to demarcate fully camera has been proposed, realize multiobject tracking based on the homography between polyphaser restriction.

But said method utilizes mixed Gauss model that background is carried out modeling, and through the problem of regular meeting generation cavity and shade, the failure of target monitoring bring very big influence can for follow-up tracking.This method need be calculated each camera, and latent disappearing just can obtain homography matrix between various visual angles, the latent calculating that disappears a little is the process of a more complicated, and the latent error of calculation that disappears a little will cause the miscount of homography matrix, and then causes the failure to multiple target tracking.The another one problem is that this method is utilized figure hugger opinion that the locating information in 10 layers in space is handled to realize following the tracks of, and causes this method at all can't requirement of real time.These three significant disadvantage have limited its application in the middle of engineering greatly.

Summary of the invention

A kind of multiple goal positioning and tracing method based on various visual angles comprises the steps:

At first a plurality of cameras are installed and delimited the common monitoring zone of described a plurality of cameras, demarcate a plurality of height layers in various visual angles;

The foreground extraction step adopts the code book model that the video image of gathering is carried out background modeling, adopts the background subtraction method to obtain the prospect likelihood image of each visual angle video image;

The homography matrix calculation procedure in conjunction with the center position of a plurality of marks on the differing heights layer of described demarcation, is calculated the homography matrix between the various visual angles on the differing heights layer of demarcating;

Prospect likelihood fusion steps, a visual angle in the selected various visual angles, as the reference visual angle, utilize the homography matrix of spending layer between described various visual angles based on each, the prospect likelihood image at other visual angle that described foreground extraction step is extracted, be mapped in the reference viewing angle, obtain the prospect likelihood fused images at a plurality of visual angles;

The multilayer fusion steps, extraction obtains through prospect likelihood fusion steps, based on the locating information of selected a plurality of height layers, utilize shortest path first, handle the locating information of each layer, obtain the multilayer pursuit path, in conjunction with the result of prospect monitoring, finish multiobject three-dimensional tracking.

Background subtraction method in the described foreground extraction step, the operating process of background subtraction method is as follows:

In the objective definition monitor procedure, newly importing pixel is x _t=(B), its corresponding code book is M for R, G,

Step 1 is calculated the brightness I=R+G+B of current pixel, definition Boolean variable match=0, and give threshold value variable ε assignment;

Step 2 finds corresponding code word C in code book M _m, if can find corresponding code word C _mThen be judged to be background image, reduced, can find corresponding codewords C _mCriterion as follows:

C. pixel x _tWith the color similarity degree of certain code word greater than detection threshold ε

The color similarity degree is defined as colordist (x _t, v _m), for the new constantly pixel x that imports of t _t

$\mathrm{colordist}(x_t,v_m)=\frac{{\left|\right|x_t\left|\right|}^2{\left|\right|v_m\left|\right|}^2-{<x_t,v_m>}^2}{{\left|\right|v_m\left|\right|}^2}$

Wherein || x _t|| ²=R ²+ G ²+ B ²,

Wherein the i value is 1,2 ... N, R, G, B are the corresponding value in R, G in the video, the B passage,

For all over after getting the i value, the mean value of corresponding R, G, B passage;

D. pixel x _tBrightness in the brightness range of this code word

Brightness changes in the moving target monitoring a scope, and for each code word, its scope is Wherein

Be respectively minimum value and maximal value that brightness changes.

In the described homography matrix calculation step, homography matrix is defined as follows:

From a described N camera, the video image that any 2 cameras are taken is designated as I respectively _i(i=1,2 ... N) and I _j(j=1,2 ... N) in order to guarantee the existence of homography, two cameras must be taken same zone on the reference planes, make X be on the π of plane more arbitrarily, X is at I _iAnd I _jIn picture m respectively _k=(x _k, y _k) and m ' _k=(x ' _k, y ' _k), k=1,2 ..m * n, m * n are the resolution of each visual angle capture video.Define one 3 * 3 matrix:

$H_{i_{\mathrm{\&pi;}}j}=\left[\begin{array}{ccc}{h}_{11}& h_{12}& h_{13}\\ h_{21}& h_{22}& h_{23}\\ h_{31}& h_{32}& 1\end{array}\right]$

Make

Matrix

Be called the homography matrix between two cameras that plane π induces, abbreviate the homography matrix of plane π as, corresponding conversion is called the homography conversion, utilizes homography matrix

Can obtain another as the corresponding point on the plane, homography matrix from one as the point on the plane

Be a homogeneous invertible matrix, have 8 degree of freedom.

For other selected height aspect of removing reference planes, the homography conversion process is as follows:

Be made as φ and be a plane in 2 planes that are parallel to reference planes, I _iPicture plane for camera i.By plane φ induce any two the picture planar I _iAnd I _jBetween homography matrix be designated as Described homography matrix

Have 8 degree of freedom, need 4 pairs of character pair points, matrix

As follows

$H_{i_{\mathrm{\&phi;}}j}=\left[\begin{array}{ccc}{h}_{11}^{\&prime;}& h_{12}^{\&prime;}& h_{13}^{\&prime;}\\ h_{21}^{\&prime;}& h_{22}^{\&prime;}& h_{23}^{\&prime;}\\ h_{31}^{\&prime;}& h_{32}^{\&prime;}& 1\end{array}\right]$

$m_k^{\&prime;}=H_{i_{\mathrm{\&phi;}}j}{m}_k,$ That is: $\left[\begin{array}{c}{x}_k^{\&prime;}\\ y_k^{\&prime;}\\ 1\end{array}\right]=\left[\begin{array}{ccc}{h}_{11}^{\&prime;}& h_{12}^{\&prime;}& h_{13}^{\&prime;}\\ h_{21}^{\&prime;}& h_{22}^{\&prime;}& h_{23}^{\&prime;}\\ h_{31}^{\&prime;}& h_{32}^{\&prime;}& 1\end{array}\right]\left[\begin{array}{c}{x}_k\\ y_{\mathrm{<figure-callout\; id="1"\; label="k"\; filenames="HDA0000059735320000011.png,HDA0000059735320000012.png"\; state="\{\{state\}\}">k</figure-callout>}}\\ 1\end{array}\right]$

Described multilayer fusion steps has step by step following:

A. on single height layer, utilize the connected region monitoring, obtain a plurality of locating information agglomerates of present frame, and calculate the distance of barycenter and each barycenter of former frame of described each agglomerate, get the target agglomerate of the shortest agglomerate of distance as coupling;

B. contrast the area of different agglomerates on described selected each height layer, choose the agglomerate of area maximum, as the elevation information that the length and width information of the three-dimensional frame of video frequency tracking is write down in conjunction with the foreground extraction unit, generate three-dimensional frame, the lock onto target of target, finish tracking to target.

Be a plurality of cameras of various visual angles distribution and the common monitoring zone of described a plurality of cameras, described common monitoring zone has selected a plurality of height layers;

Foreground extracting module adopts the code book model that the video image of gathering is carried out background modeling, adopts the background subtraction method to obtain the prospect likelihood image of each visual angle video image;

The homography matrix computing module in conjunction with the center position of a plurality of marks on the differing heights layer of described demarcation, calculates the homography matrix between each visual angle on the differing heights layer of demarcating;

Prospect likelihood Fusion Module, a visual angle in the selected various visual angles, as the reference visual angle, utilize between described various visual angles homography matrix based on multilayer, the prospect likelihood image at other visual angle that described foreground extraction step is extracted, be mapped in the reference viewing angle, obtain the prospect likelihood fused images at a plurality of visual angles;

The multilayer Fusion Module, extraction obtains through prospect likelihood fusion steps, based on the locating information of selected a plurality of height layers, utilizes shortest path first, handles the locating information of each layer;

Tracking module obtains the multilayer pursuit path, in conjunction with the result of prospect monitoring, finishes multiobject three-dimensional tracking.

Described prospect likelihood integrated unit, coordinate in conjunction with the many marks central point that is positioned at the differing heights layer, calculate the homography matrix between each visual angle on the differing heights layer, an equipment in a plurality of video monitoring equipments of selected described formation various visual angles is as the reference visual angle, according between the described various visual angles that calculate based on the homography matrix of each aspect, prospect likelihood image with other visual angle is mapped in the reference viewing angle, obtains the prospect likelihood fused images at a plurality of visual angles.

This method does not need to calculate the latent of polyphaser and disappears a little in tracing process, greatly reduce the complexity of calculating, has improved the accuracy of homography matrix simultaneously yet.Introduce the code book model first and solve the multiple target tracking problem, improved the accuracy of following the tracks of.This method positions multiple goal perpendicular to three layers of ground in the space, utilizes shortest path first's algorithm process multilayer locating information, realizes multiobject tracking, has significantly improved the computing velocity of this method.The monocular track algorithm can't obtain Three-dimension Target information, utilizes the information fusion between polyphaser can realize following the tracks of multiple goal with three-dimensional frame.In sum, the present invention proposes to have good stability based on the multi-object tracking method of polyphaser, and real-time is good, the characteristics that precision is high.

Description of drawings

Fig. 1 is a structure diagram of the present invention,

Fig. 2 is the structure diagram of video image handling part of the present invention,

Fig. 3 is a process flow diagram of the present invention,

Among the figure: 1. video image acquisition unit, 2 Video processing portions, 3 storage parts, 201. foreground extraction unit, 202 prospect likelihood integrated units, 203. multilayer integrated units, 204. tracking cell.

Embodiment

Fig. 1 is a structured flowchart of the present invention, as shown in Figure 1:

A kind of multiple goal locating and tracking system based on polyphaser, comprise image acquiring unit 1, comprise the video monitoring regional of a demarcation, the common shooting area of a plurality of cameras promptly is set, choose parallel differing heights on the direction perpendicular to ground three layers, on each layer, place mark post.Concrete performing step is as follows: place 4 mark posts perpendicular to ground in photographed scene, on identical height, as a preferred implementation, mark on mark post with red, be convenient to video equipment identification.

Storage part 3, the data that arrive in order to storage of collected.

And a plurality of cameras a plurality of dispersions, that be a plurality of different visual angles, in order to guarantee the accuracy of video monitoring, the present invention adopts the video monitoring equipment more than three.

Fig. 2 is the structured flowchart of invention Video processing portion, as shown in Figure 2:

Video processing of the present invention portion comprises 2:

Foreground extraction unit 201 is handled the data that the collection of video image acquisition unit comes, and isolates the prospect likelihood image that each video monitoring equipment collects, and with the prospect likelihood image that extracts, is sent to prospect likelihood integrated unit 202.

Described prospect likelihood integrated unit 202, main homography matrix and selected reference viewing angle of being responsible between each video monitoring equipment of computing, on each selected height aspect, the prospect likelihood image that other visual angle is collected is according to this homography matrix, be mapped to reference viewing angle, also adopt same processing for other selected in advance two height layers.

Described multilayer integrated unit 203 extracts the prospect likelihood fused images based on selected differing heights layer, utilizes shortest path first to handle the multilayer locating information,

Tracking module 204 obtains the multilayer pursuit path, in conjunction with the result of foreground detection, finishes multiobject three-dimensional tracking.

Fig. 3 is a process flow diagram of the present invention

A kind of as shown in Figure 3 multiple goal positioning and tracing method based on various visual angles has following steps:

Image acquisition step comprises the video monitoring regional of demarcation a plurality of cameras being set promptly, at least three cameras, common shooting area, three layers of choosing the differing heights that is parallel to ground on the direction perpendicular to ground.Concrete performing step is as follows: place 4 mark posts perpendicular to ground in photographed scene, on selected identical height, as a preferred implementation, mark on mark post with red, be convenient to video equipment identification.

The foreground extraction step adopts the code book model that the video image of gathering is carried out background modeling, adopts the background subtraction method to obtain the prospect likelihood image of each visual angle video image; The method of following list of references record is used in the foundation of described code book model, repeats no more here.

1 one kinds of monitor video moving target policing algorithm periodicals " computer engineering " 2007/07/20 of documents based on code book

Documents 2Kim K, Chalidabhongse T H, Harwood D, Davi s L.Real-Time foreground-background segmentation using codebook model[J], Real-Time Imaging, 2005,11 (3): 167-256.

The concrete steps of described background subtraction operation are as follows

New input pixel is x in the hypothetical target monitor procedure _t=(B), its corresponding code book is M for R, G, as a preferred implementation, and background subtraction operation BGS (x _t) be divided into for three steps:

Step 1. is calculated the brightness I=R+G+B of current pixel, and wherein R, G, B are red in the video, green and blue brightness value.Definition Boolean variable match=0, and give threshold value variable ε assignment.

Step 2. is found out the code word C that is complementary with current pixel according to following two conditions from code book M _mIf can find code word C _m, match=1 then, otherwise match=0 judges current pixel and code word C _mWhether mate, need satisfy following 2 conditions:

Condition 1

The color similarity degree is defined as colordist (x _t, v _m), for the new constantly pixel x that imports of t _t

$\mathrm{colordist}(x_t,v_m)=\frac{{\left|\right|x_t\left|\right|}^2{\left|\right|v_m\left|\right|}^2-{<x_t,v_m>}^2}{{\left|\right|v_m\left|\right|}^2}$

Wherein || x _t|| ²=R ²+ G ²+ B ²,

Wherein the i value is 1,2 ... N, R, G, B are the corresponding value in R, G in the video, the B passage,

For all over after getting the i value, the mean value of corresponding R, G, B passage;

Condition 2:

For judgement prospect and background, brightness changes in the moving target monitoring a scope, for each code word, when

The time, Wherein:

In the following formula

Be the code word C in the corresponding code book _mBrightness value, I _Hi, I _LowBrightness range for whole code book.The span 0.4-0.7 of α, the span of β is 1.1-1.5

Step 3. is judged the foreground moving object pixel:

$\mathrm{BGS}\left(x_t\right)=\left\{\begin{array}{cc}\mathrm{foreground}& \mathrm{match}=0\\ \mathrm{background}& \mathrm{match}=1\end{array}\right.$

Foreground is that foreground elements, background are background element.

If match=0 judges that then object pixel is a prospect, it is background element that match=1 then is judged to be object pixel, and is reduced, if can not find corresponding code word in code book, then thinks foreground pixel, is kept.

Described homography matrix calculation step

The calculation step of homography matrix is described, the notion of the homography matrix between various visual angles is: establishing π is the georeferencing plane of not passing through two arbitrary photocentres of camera, selectively the plane is as the reference plane in experiment for we, and the image of two camera shootings (being called for short as the plane) is designated as I respectively _iAnd I _j. in order to guarantee the existence of homography, two cameras must be taken same zone on the reference planes. make X be on the π of plane more arbitrarily, X is at I _iAnd I _jIn picture m respectively _k=(x _k, y _k) and m ' _k=(x ' _k, y ' _k), the span of i, j is 1,2 here ..., N, N are the number of camera, k=1, and 2 ..m * n, m * n are the resolution of each visual angle capture video.We define one 3 * 3 matrix:

$H_{i_{\mathrm{\&pi;}}j}=\left[\begin{array}{ccc}{h}_{11}& h_{12}& h_{13}\\ h_{21}& h_{22}& h_{23}\\ h_{31}& h_{32}& 1\end{array}\right]$

Make

Satisfy the matrix of following formula Be called the homography matrix of (or two picture interplanars) between two cameras that plane π induces, or abbreviate the homography matrix of plane π as, corresponding conversion is called the homography conversion.Utilize homography matrix

Can obtain another as the corresponding point on the plane from one as the point on the plane.Homography matrix Be a homogeneous invertible matrix, have 8 degree of freedom.

As a preferred implementation, 3 layers of the height layer positions of choosing among the present invention, it is a plane that is parallel in 2 planes of reference planes that height layer is established φ, I _iPicture plane for camera i.By plane φ induce any two the picture planar I _iAnd I _jBetween homography matrix be designated as

We as can be seen

Have 8 degree of freedom,, just can calculate homography matrix based on each layer if we can find 4 pairs of characteristic of correspondence points on each layer.In experimentation, we place 4 poles in the environment of taking, 4 mark posts are set respectively on 3 layers of differing heights, utilize the center of mark post in image to calculate the homography matrix of each layer.Through experimental demonstration, this method has been avoided the latent complicated processes that disappears a little of monitoring, has improved calculating

Accuracy, Matrix is as follows:

$H_{i_{\mathrm{\&phi;}}j}=\left[\begin{array}{ccc}{h}_{11}^{\&prime;}& h_{12}^{\&prime;}& h_{13}^{\&prime;}\\ h_{21}^{\&prime;}& h_{22}^{\&prime;}& h_{23}^{\&prime;}\\ h_{31}^{\&prime;}& h_{32}^{\&prime;}& 1\end{array}\right]$

Make Set up, that is: $\left[\begin{array}{c}{x}_k^{\&prime;}\\ y_k^{\&prime;}\\ 1\end{array}\right]=\left[\begin{array}{ccc}{h}_{11}^{\&prime;}& h_{12}^{\&prime;}& h_{13}^{\&prime;}\\ h_{21}^{\&prime;}& h_{22}^{\&prime;}& h_{23}^{\&prime;}\\ h_{31}^{\&prime;}& h_{32}^{\&prime;}& 1\end{array}\right]\left[\begin{array}{c}{x}_{\mathrm{<figure-callout\; id="1"\; label="k\; y\; k"\; filenames="HDA0000059735320000011.png,HDA0000059735320000012.png"\; state="\{\{state\}\}">k</figure-callout>}}& \\ y_k{}_{}\\ 1\end{array}\right].$

After calculating the homography matrix between various visual angles, the step that the various visual angles likelihood is merged is as follows:

Suppose to choose N camera, the visual angle of selecting j camera based on the homography mapping relations of multilayer, utilizes formula as the reference visual angle between the utilization various visual angles

Prospect likelihood information L with other N-1 visual angle _i(m _k) being mapped to reference viewing angle j, the foreground likelihood function after obtaining shining upon is for being designated as

Prospect likelihood image with N visual angle merges in reference viewing angle then, and the likelihood function of the image after the fusion is:

$L_{\mathrm{\&Sigma;}}\left(m_k^{\&prime;}\right)=\frac{1}{N}\{L_j\left(m_k^{\&prime;}\right)+\underset{i=1}{\overset{N-1}{\mathrm{\&Sigma;}}}{L}_{i_{\mathrm{\&phi;}}j}\left(m_k^{\&prime;}\right)\}$

(i in the formula, the j value is 1,2 ... N, N are the camera number; K=1,2 ..m * n)

Merge figure based on the prospect likelihood of multilayer between promptly obtaining from various visual angles, the high bright spot in the fused images on each layer is the locating information of multiple goal on multilayer.

Described fusion steps is extracted the locating information based on selected height layer, utilizes shortest path first's algorithm process multilayer locating information.Concrete process is: after obtaining the fusion figure of reference viewing angle, we have obtained in everyone positional information on each layer, because we use a plurality of video cameras, therefore guaranteeing that homography matrix is accurately under the situation, the fusion figure of reference viewing angle has made equivalent foreground segmentation, the possibility of result that obtains is understood the noise point of some, utilizes the morphology opening and closing operations to do pre-service earlier.Processing procedure with one deck locating information is an example, and other two-layer locating information processing procedures are identical.The process of shortest path first's algorithm is as follows: any frame in the selected video image utilizes the connected region monitoring can obtain the locating information agglomerate R of this frame target i as present frame _i, calculate R _iBarycenter

The horizontal ordinate of barycenter is respectively:

$\stackrel{\&OverBar;}{x}=\frac{1}{A}\underset{(x,y){\&Element;R}_i}{\mathrm{\&Sigma;}}x$

$\stackrel{\&OverBar;}{y}=\frac{1}{A}\underset{(x,y){\&Element;R}_i}{\mathrm{\&Sigma;}}y$

Wherein A is agglomerate R _iArea.Next, calculate the distance at each barycenter of present frame and each center of former frame

Get the target agglomerate of the agglomerate of centroid distance minimum as coupling.

Contrast the area of different agglomerates on described selected each height layer, choose the agglomerate of area maximum, as the elevation information that the length and width information of the three-dimensional frame of video frequency tracking is write down in conjunction with the foreground extraction unit, generate three-dimensional frame, the lock onto target of target, finish tracking to target.

The above; only be the preferable embodiment of the present invention; but protection scope of the present invention is not limited thereto; anyly be familiar with those skilled in the art in the technical scope that the present invention discloses; be equal to replacement or change according to technical scheme of the present invention and inventive concept thereof, all should be encompassed within protection scope of the present invention.

Claims (6)

1. the multiple goal positioning and tracing method based on polyphaser is characterized in that comprising the steps:

At first a plurality of cameras are installed, delimited the common monitoring zone of described a plurality of cameras, demarcate a plurality of height layers in various visual angles;

The foreground extraction step adopts the code book model that the video image of gathering is carried out background modeling, adopts the background subtraction method to obtain the prospect likelihood image of each visual angle video image;

The homography matrix calculation procedure in conjunction with the center position of a plurality of marks on the differing heights layer of described demarcation, is calculated the homography matrix between the various visual angles on the differing heights layer of demarcating;

Prospect likelihood fusion steps, a visual angle in the selected various visual angles is as the reference visual angle, utilize the homography matrix of spending layer between described various visual angles based on each, the prospect likelihood image at other visual angle that described foreground extraction step is extracted, be mapped in the reference viewing angle, obtain the prospect likelihood fused images at a plurality of visual angles;

The multilayer fusion steps, extraction obtains through prospect likelihood fusion steps, based on the locating information of selected a plurality of height layers, utilize shortest path first, handle the locating information of each layer, obtain the multilayer pursuit path, in conjunction with the result of foreground extraction, finish multiobject three-dimensional tracking.

2. the multiple goal positioning and tracing method based on various visual angles according to claim 1, its feature also is: background subtraction method in the described foreground extraction step, the operating process of background subtraction method is as follows:

In the objective definition monitor procedure, newly importing pixel is x _t=(B), its corresponding code book is M for R, G,

Step 1 is calculated the brightness I=R+G+B of current pixel, definition Boolean variable match=0, and give threshold value variable ε assignment;

Step 2 finds corresponding code word C in code book M _m, if can find corresponding code word C _mThen be judged to be background image, reduced, can find corresponding codewords C _mCriterion as follows:

A. pixel x _tWith the color similarity degree of certain code word greater than detection threshold ε

The color similarity degree is defined as colordist (x _t, v _m), for the new constantly pixel x that imports of t _t

$\mathrm{colordist}(x_t,v_m)=\frac{{\left|\right|x_t\left|\right|}^2{\left|\right|v_m\left|\right|}^2-{<x_t,v_m>}^2}{{\left|\right|v_m\left|\right|}^2}$

Wherein || x _t|| ²=R ²+ G ²+ B ²,

Wherein the i value is 1,2 ... N, R, G, B are the corresponding value in R, G in the video, the B passage,

For all over after getting the i value, the mean value of corresponding R, G, B passage;

B. pixel x _tBrightness in the brightness range of this code word

Brightness changes in the moving target monitoring a scope, and for each code word, its scope is

Wherein

Be respectively minimum value and maximal value that brightness changes.

3. the multiple goal positioning and tracing method based on various visual angles according to claim 1, its feature also is: in the described homography matrix calculation step, homography matrix is defined as:

From a described N camera, the video image that any 2 cameras are taken is designated as I respectively _i(i=1,2 ... N) and I _j(j=1,2 ... N) in order to guarantee the existence of homography, two cameras must be taken same zone on the reference planes, make X be on the π of plane more arbitrarily, X is at I _iAnd I _jIn picture m respectively _k=(x _k, y _k) and m ' _k=(x ' _k, y ' _k), k=1,2 ..m * n, m * n are the resolution of each visual angle capture video, define one 3 * 3 matrix:

$H_{i_{\mathrm{\&pi;}}j}=\left[\begin{array}{ccc}{h}_{11}& h_{12}& h_{13}\\ h_{21}& h_{22}& h_{23}\\ h_{31}& h_{32}& 1\end{array}\right]$

Make

Matrix

Be called the homography matrix between two cameras that plane π induces, abbreviate the homography matrix of plane π as, corresponding conversion is called the homography conversion, and one multiplies each other as the point on the plane and homography matrix and to utilize homography matrix

Obtain another as the corresponding point on the plane, homography matrix from one as the point on the plane

Be a homogeneous invertible matrix, have 8 degree of freedom;

For other selected height aspect of removing reference planes, the homography conversion process is as follows:

Be made as φ and be a plane in 2 planes that are parallel to reference planes, I _iPicture plane for camera i.By plane φ induce any two the picture planar I _iAnd I _jBetween homography matrix be designated as

Described homography matrix

Have 8 degree of freedom, need 4 pairs of character pair points, matrix

As follows:

$H_{i_{\mathrm{\&phi;}}j}=\left[\begin{array}{ccc}{h}_{11}^{\&prime;}& h_{12}^{\&prime;}& h_{13}^{\&prime;}\\ h_{21}^{\&prime;}& h_{22}^{\&prime;}& h_{23}^{\&prime;}\\ h_{31}^{\&prime;}& h_{32}^{\&prime;}& 1\end{array}\right]$

$m_k^{\&prime;}=H_{i_{\mathrm{\&phi;}}j}{m}_k,$ That is: $\left[\begin{array}{c}{x}_k^{\&prime;}\\ y_k^{\&prime;}\\ 1\end{array}\right]=\left[\begin{array}{ccc}{h}_{11}^{\&prime;}& h_{12}^{\&prime;}& h_{13}^{\&prime;}\\ h_{21}^{\&prime;}& h_{22}^{\&prime;}& h_{23}^{\&prime;}\\ h_{31}^{\&prime;}& h_{32}^{\&prime;}& 1\end{array}\right]\left[\begin{array}{c}{x}_k\\ y_k\\ 1\end{array}\right].$

4. the multiple goal positioning and tracing method based on various visual angles according to claim 1, its feature also is: described multilayer fusion steps has following several steps:

A. on single height layer, utilize the connected region monitoring, obtain a plurality of locating information agglomerates of present frame, and calculate the distance of barycenter and each barycenter of former frame of described each agglomerate, get the target agglomerate of the shortest agglomerate of distance as coupling;

B. contrast the area of different agglomerates on described selected each height layer, choose the agglomerate of area maximum, as the elevation information that the length and width information of the three-dimensional frame of video frequency tracking is write down in conjunction with the foreground extraction unit, generate three-dimensional frame, the lock onto target of target, finish tracking to target.

One kind based on various visual angles the multiple goal locating and tracking system, it is characterized in that having:

Be a plurality of cameras of various visual angles distribution and the common monitoring zone of described a plurality of cameras, described common monitoring zone has selected a plurality of height layers;

Foreground extracting module adopts the code book model that the video image of gathering is carried out background modeling, adopts the background subtraction method to obtain the prospect likelihood image of each visual angle video image;

The homography matrix computing module in conjunction with the center position of a plurality of marks on the differing heights layer of described demarcation, calculates the homography matrix between each visual angle on the differing heights layer of demarcating;

Prospect likelihood Fusion Module, a visual angle in the selected various visual angles, as the reference visual angle, the prospect likelihood image at other visual angle that described foreground extraction step is extracted is mapped in the reference viewing angle, obtains the prospect likelihood fused images at a plurality of visual angles;

The multilayer Fusion Module, extraction obtains through prospect likelihood fusion steps, based on the locating information of selected a plurality of height layers, utilizes shortest path first, handles the locating information of each layer;

Tracking module obtains the multilayer pursuit path, in conjunction with the result of prospect monitoring, finishes multiobject three-dimensional tracking.

6. the multiple goal locating and tracking system based on polyphaser according to claim 5, its feature also is: described prospect likelihood integrated unit, coordinate in conjunction with the many marks central point that is positioned at the differing heights layer, calculate the homography matrix between each visual angle on the differing heights layer, an equipment in a plurality of video monitoring equipments of selected described formation various visual angles is as the reference visual angle, according between the described various visual angles that calculate based on the homography matrix of each aspect, prospect likelihood image with other visual angle, be mapped in the reference viewing angle, obtain the prospect likelihood fused images at a plurality of visual angles.

Images