CN104063883A - Surveillance video abstract generating method based on combination of object and key frames - Google Patents

Abstract

The invention relates to the field of video image processing, in particular to a surveillance video abstract generating method based on a combination of an object and key frames. The surveillance video abstract generating method includes the steps of extracting movement fragments, setting up a mixed Gaussian background model, detecting the movement object in a background deduction mode, tracking the movement object based on Kalman prediction, extracting the key frames after image frames are accurately clustered through preliminary extraction and further classification, deducting the movement object and synthesizing an abstract. The surveillance video abstract generating method has the advantages that a main algorithm flow of the video abstract based on the object is adopted, an improved key frame extracting method is integrated, the key frames are extracted after the image frames are accurately clustered through preliminary extraction and further classification, video redundant information is reduced as far as possible on the premise of not leaving out key information, and the compression degree of videos is good; meanwhile, a reasonable video synthesis step is provided so that a user can read the surveillance content of the videos in a short time and position useful information rapidly and effectively.

Description

A kind of monitor video abstraction generating method combining based on object and key frame

Technical field

The present invention relates to field of video image processing, relate in particular to a kind of monitor video abstraction generating method combining based on object and key frame.

Background technology

Along with the development of digital video technology, increasing monitoring video equipment is deployed in each place of society, various monitoring video device real-time monitorings, all collect the video information of magnanimity every day, therefore the requirement of video analysis and processing are also improved constantly.General monitoring video equipment all needs monitor staff to monitor video, but how fast browsing video data from the video information of magnanimity, accurately, certain time period occurrence of fast finding, a job extremely taking time and effort for monitor staff, and for monitor staff, often to pay close attention to many watch-dogs simultaneously, may not be concerned about in real time each monitoring details, be easy to omit because of the absent minded of personnel.

In order to help monitor staff's fast browsing video data, convenient inquiry video information, the analysis based on video content has proposed video summarization method.Video frequency abstract claims again video concentrated, to video content simplified summary, by to moving target analysis, extract moving target, then the movement locus of each target is analyzed, different targets is spliced in a common background scene, and they are combined in some way and obtain new video content.The research of existing video frequency abstract is mainly with separately based on key frame be basic based on object separately.The keyframe sequence that video summarization method based on key frame is chosen in former video combines video intentionally, and the method exists shortcoming that can not expressed intact video semanteme, causes multitude of video information loss, cannot Overall Acquisition video information; Although object-based video summarization method can by Moving Objects track give depicted, this method video compress degree is not strong.

Summary of the invention

The present invention overcomes above-mentioned weak point, and object is to provide a kind of monitor video abstraction generating method combining based on object and key frame, and the method can solve the problem that existing video frequency abstract video semanteme is imperfect, compression degree is not strong.

The present invention achieves the above object by the following technical programs: a kind of monitor video abstraction generating method combining based on object and key frame, comprises the following steps:

1) adjacent two two field pictures are contrasted and obtain changing value, changing value and threshold value compare, and the image that reservation changing value is greater than threshold value is motion fragment, and motion fragment forms new video;

2) new video is evenly divided into several fragments according to video time, utilizes mixture Gaussian background model to the picture frame production background model in each fragment, background model adds that timestamp information deposits in background database;

3) any t moment, current frame image I _twith background model B _trelatively obtain the difference value at (x, y) some place, difference value and threshold value T comparison, the part that difference value is greater than T is Moving Objects, all the other are background;

4) based on Kalman filtering algorithm, Moving Objects is followed the tracks of, the trace information of all Moving Objects deposits in object database;

5) the same Moving Objects detecting is marked, utilize the color characteristic of object to carry out similarity contrast, sort out for the picture frame that similarity is high, to of a sort picture frame calculating mean value, choosing with the frame of mean value difference minimum is key frame;

6) adopt GraphCut algorithm to carry out Moving Objects to take off to marking the region of Moving Objects, from background database, take out the background image of different time sections and copy, the Moving Objects of taking off is put in chronological order successively to the correspondence position of background image, formed video frequency abstract.

As preferably, described step 5) in, taking the first frame as reference frame, its color histogram is V ₁, the color histogram of subsequent frame is V _i, V _iwith V ₁similarity than for S _i, every frame is divided into MxN piece, and computing formula is:

$S_i=\underset{j=1}{\overset{M\×N}{\mathrm{\Σ}}}{S}_{\mathrm{ij}}{M}_j/M\×N$

S _ij＝|V _ij-V _1j|

Wherein, S _ijthe similarity value of i frame j piece, M _jit is the weight factor of j piece;

According to above-mentioned formula, picture frame is sorted out, if S _ibe less than or equal to threshold alpha, picture frame and V ₁be classified as a class picture frame L ₁if, α <S _i≤ α ₀, picture frame is classified as and treats framing, if α is >S ₀, picture frame is classified as Equations of The Second Kind L ₂; Then with Equations of The Second Kind L ₂the first frame as reference frame, carry out similarity contrast with each frame below, complete the contrast of all picture frames.

As preferably, the described framing for the treatment of is set up to worker thread, adopt the color similarity contrast of channel separation, specifically comprise the following steps:

To input reference frame image and two field picture undetermined utilize cvSplit function by the separation of images of rgb format the image for B, G, tri-passages of R, bImg, gImg, rImg, three of reference frame image and two field picture undetermined Color Channels are carried out respectively to similarity contrast, if the similarity of three passages all in threshold value two field picture undetermined and reference frame image be same class, otherwise treat that framing is classified as next class.

As preferably, described same Moving Objects adopts minimum rectangle frame to mark, and rectangle frame is that similarity comparison domain and Moving Objects are taken off region.

As preferably, described step 6) specifically comprise the following steps:

6.1) take out the background image frame obtaining in first time period;

6.2) all Moving Objects of taking off in this time period are placed and numbered at the correspondence position of background image in chronological order;

6.3) object part overlapped in a two field picture is carried out to translucentization processing;

6.4) in the time that being greater than certain threshold value, the object number of placing in a two field picture stops placing;

6.5) copy background image frame and continue placing objects until discharge;

6.6) take out the background image frame in the next time period and the object of taking off in this time period, repeating step 6.2)-6.5);

6.7) all picture frames after synthetic are combined, form final video frequency abstract.

Beneficial effect of the present invention is: the inventive method adopts a kind of main algorithm flow process of object-based video frequency abstract, and incorporate improved extraction method of key frame, to after accurate picture frame cluster, extract key frame by preliminary extraction and further classification, under the prerequisite of not omitting key message, reduce as much as possible video redundant information, video compress degree is better; Propose comparatively reasonably video synthesis step of one simultaneously, made user can read at short notice video monitoring content, fast and effeciently located useful information.

Brief description of the drawings

Fig. 1 is method step process flow diagram of the present invention;

Fig. 2 is the flow chart of steps of key-frame extraction;

Fig. 3 is that in the embodiment of the present invention, the weights of picture frame are cut apart figure.

Embodiment

Below in conjunction with specific embodiment, the present invention is described further, but protection scope of the present invention is not limited in this:

Embodiment 1: as shown in Figure 1, a kind of monitor video abstraction generating method combining based on object and key frame, it comprises the following steps:

(1) extract motion fragment:

Adopt the method for consecutive frame contrast continuously adjacent two frames to be contrasted, if change and be greater than threshold value between two frames, regarded as Moving Objects and occurred, be extracted as motion fragment, otherwise give up to fall as non-motion fragment.

Taking the picture frame of m × n size as example, the gray-scale value of establishing each pixel is G _ij, the pixel matrix of K two field picture is:

$G^K=\left(\begin{array}{cccc}{G^K}_{11}& {G^K}_{12}& \&CenterDot;\&CenterDot;\&CenterDot;& {G^K}_{1n}\\ {G^K}_{21}& {G^K}_{22}& \&CenterDot;\&CenterDot;\&CenterDot;& {G^K}_{2n}\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ {G^K}_{m1}& {G^K}_{m2}& \&CenterDot;\&CenterDot;\&CenterDot;& {G^K}_{\mathrm{mn}}\end{array}\right)$

If m row row vector A=(11 ... 1), n ranks vector B=(11 ... 1) ^t,

$C^K=A\&times;G^K=(\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{G_{i1}}^K,\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{G_{i2}}^K,\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{G_{i3}}^K\&CenterDot;\&CenterDot;\&CenterDot;\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{G_{\mathrm{in}}}^K)---\left(1\right)$

$R^K=G^K\&times;B={(\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{G_{1j}}^K,\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{G_{2j}}^K,\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{G_{3j}}^K\&CenterDot;\&CenterDot;\&CenterDot;\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{G_{\mathrm{mj}}}^K)}^T---\left(2\right)$

In like manner can obtain the C of K+1 frame ^k+1and R ^k+1be respectively:

$C^{K+1}=A\&times;G^{K+1}=(\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{G_{i1}}^{K+1},\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{G_{i2}}^{K+1},\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{G_{i3}}^{K+1}\&CenterDot;\&CenterDot;\&CenterDot;\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{G_{\mathrm{in}}}^{K+1})---\left(3\right)$

$R^{K+1}=G^{K+1}\&times;B={(\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{G_{1j}}^{K+1},\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{G_{2j}}^{K+1},\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{G_{3j}}^{K+1}\&CenterDot;\&CenterDot;\&CenterDot;\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{G_{\mathrm{mj}}}^{K+1})}^T---\left(4\right)$

The corresponding row column position of K frame and K+1 frame is subtracted each other and is taken absolute value:

$C_d=\left(\right|C^{K+1}-C^K\left|\right)=\left(\right|\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{G_{i1}}^{K+1}-\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{G_{i1}}^K|,|\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{G_{i2}}^{K+1}-\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{G_{i2}}^K|,\&CenterDot;\&CenterDot;\&CenterDot;,|\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{G_{\mathrm{in}}}^{K+1}-\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{G_{\mathrm{in}}}^K\left|\right)---\left(5\right)$

$R_d=\left(\right|R^{K+1}-R^K\left|\right)={\left(\right|\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{G_{\mathrm{Ij}}}^{K+1}-\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{G_{1j}}^K|,|\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{G_{2j}}^{K+1}-\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{G_{2j}}^K|,\&CenterDot;\&CenterDot;\&CenterDot;,|\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{G_{\mathrm{mj}}}^{K+1}-\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{G_{\mathrm{mj}}}^K\left|\right)}^T---\left(5\right)$

Order

Σ _C＝C _d×B

Σ _R＝A×R _d

Finally draw two value Σ that represent respectively row and column contrast _cand Σ _r, then respectively with threshold value T _c, T _rcontrast, if meet

Σ _C＞T _C||Σ _R≤T _R(7)

The correlative value that is row and column has a threshold value that exceedes setting, is judged as motion fragment, otherwise is non-motion fragment, gives up non-motion fragment to reduce redundancy video.

(2) mixed Gaussian background modeling:

Each pixel is set up to K Gauss model and reflect time dependent real background, new picture frame is mated, and continuous Renewal model.Each pixel is set up to mixed Gauss model as follows:

$P\left(x_t\right)=\underset{i=1}{\overset{K}{\mathrm{\&Sigma;}}}{w}_{i,t}*\mathrm{\&eta;}(x_t,u_{i,t},{\mathrm{\&Sigma;}}_{i,t})---\left(8\right)$

$\mathrm{\&eta;}(x_t,u_{i,t},{\mathrm{\&Sigma;}}_{i,t})=\frac{1}{{\left(2\mathrm{\&pi;}\right)}^{\frac{n}{2}}{\left|\mathrm{\&Sigma;}\right|}^{\frac{1}{2}}}{e}^{-\frac{1}{2}{(x_t-u_t)}^T{\mathrm{\&Sigma;}}^{-1}(x_t-u_t)}---\left(\text{9}\right)$

Wherein w _i,tbe i the weights that Gauss model is corresponding, η (x _t, u _i,t, Σ _i,t) be the probability density function of i Gauss model, Σ _i,tit is the covariance matrix of i Gauss model.

The weight w of t moment i Gauss model _i,tmore new formula is as follows:

w _i,t＝(1-α)w _i,t-1+α(M _i,t) (10)

Wherein α is learning rate, M _i,tfor matching degree, be 1 to the model value of coupling, remaining value is 0, constant to unmatched model maintenance, the model modification of coupling is as follows:

u _t＝(1-ρ)u _t-1+ρx _t(11)

σ ² _t＝(1-ρ)σ ² _t-1+ρ(x _t-u _t) ^T(x _t-u _t) (12)

ρ＝αη(x _t|u _k,σ _k) (13)

All Gaussian distribution order is according to priority arranged, and the top n of getting wherein distributes according to weight generation background model, and the selection rule of N is:

N＝argmin _n(w _k＞T) (14)

Wherein T is predefined threshold value.

The mode that adopts timing extraction background model, is divided into N time period uniformly by video time, the picture frame in each time period is adopted to the method generation background model of mixed Gauss model, adds that timestamp information deposits background database in.

(3) background subtraction detects Moving Objects:

Can obtain the background image Model B of any moment t according to background modeling above _t, current frame image I _twith background image in the gray-scale value difference result at (x, y) some place be:

Δd _t(x,y)＝|I _t(x,y)-B _t(x,y)| (15)

Wherein Δ d _t(x, y), I _t(x, y), B _t(x, y) represents respectively the gray-scale value that difference image, current frame image, the background image in t moment are located at (x, y), and by difference image gray-scale value and threshold value T contrast, the part that is greater than T is defined as Moving Objects part, otherwise is background parts.

$d_t(x,y)=\left\{\begin{array}{cc}0,& d_t(x,y)<T\\ 1,& d_t(x,y)\&GreaterEqual;T\end{array}\right.---\left(16\right)$

(4) moving object tracking based on Kalman prediction

First adopt the position that in Kalman filtering algorithm prediction next frame image, Moving Objects occurs, obtain the center point coordinate of forecasting object rectangle frame, then calculate the two-dimensional distance between the actual object rectangle frame center point coordinate detecting, if distance is less than threshold value, think that the match is successful, and upgating object, if mate unsuccessful, think that former object disappears, and new object is followed the tracks of, the Moving Objects of each tracking is added to timestamp information deposits object database in.

Kalman filtering algorithm flow process is as follows:

The predictive equation of state vector:

$\hat{X}(k,k-1)=A(k,k-1)\hat{X}(k-1)---\left(17\right)$

The prediction variance of error covariance:

p(k,k-1)＝A(k,k-1)P(k-1)A ^T(k,k-1)+Q(k-1) (18)

The gain of Kalman filtering:

K (k)=P (k, k-1) H ^t(k) (H (k) P (k, k-1) H ^t(k)+R (k)) ^-1(19) to state vector correction:

$\hat{X}\left(k\right)=\hat{X}(k,k-1)+K\left(k\right)(Z\left(k\right)-H\left(k\right)\hat{X}(k,k-1))---\left(20\right)$

To the correction of error covariance matrix:

P(k)＝P(k,k-1)-K(k)H(k)P(k,k-1) (21)

If i the object centers point coordinate obtaining by Kalman filter forecasting is j the object centers point coordinate detecting is X _j(x _{j, n+1}, y _{j, n+1}), the two-dimensional distance recording is:

$D_{i,j}=\sqrt{{({\hat{x}}_{i,n+1}-x_{j,n+1})}^2+{({\hat{y}}_{i,n+1}-y_{j,n+1})}^2}---\left(22\right)$

Judged result after binaryzation is:

$X_j=\left\{\begin{array}{cc}1,& D_{i,j}\&le;T\\ 0,& D_{i,j}>T\end{array}\right.---\left(23\right)$

(5) key-frame extraction, as shown in Figure 2:

The present invention adopts improved extraction method of key frame, the method contrasting by weighting in system main thread is carried out preliminary extraction, set up worker thread simultaneously, do not affecting under the prerequisite of system real time, the method that adopts split tunnel to contrast respectively is further differentiated.

1) weighting is tentatively extracted:

For the same Moving Objects detecting, all with minimum rectangle collimation mark note, taking its rectangle frame as matching area, the pixel in different images frame is carried out to similarity Characteristic Contrast at each frame.

Utilize the color characteristic of object to carry out similarity contrast, adopt the color characteristic of color histogram V presentation graphs picture frame, carry out the similarity between indicated object with S (Similar writes a Chinese character in simplified form).

Taking the first frame as reference frame, its color histogram is V ₁, the color histogram V of subsequent frame _irespectively with V ₁carry out similarity contrast, result is S _iif, S _ibe less than or equal to threshold alpha, with V ₁be classified as a class picture frame L ₁if, α <S _i≤ α ₀, be classified as and treat framing, if α is >S ₀, be classified as Equations of The Second Kind L ₂, then with the first frame of Equations of The Second Kind as reference frame, carry out similarity contrast with each frame below, by that analogy.

Consider the interior different pixels point of rectangle frame shared importance degree difference in similarity contrast of object, with the matrix of a 4*4, rectangle frame is divided into 16 pieces herein, as shown in Fig. 3-(b), first each piece is carried out to similarity contrast, then participate in the similarity computing of this frame according to the weights of the each point of matrix, weights distribute as shown in Fig. 3-(a).

The similarity of i frame j piece is S _ij, can calculate respectively the similarity in each piece region and the 1st frame corresponding region in i frame, can calculate i two field picture V after adding the weight factor M of every _iwith V ₁between similarity S _i, computing formula is as follows:

S _ij＝|V _ij-V _1j| (24)

$S_i=\underset{j=1}{\overset{16}{\mathrm{\&Sigma;}}}{S}_{\mathrm{ij}}{M}_j/36---\left(25\right)$

Through the processing of above process, can complete the preliminary classification of similar frame, for being classified as the picture frame for the treatment of framing, set up worker thread, the mode that adopts channel separation and each Color Channel is carried out to similarity contrast is carried out more accurate pixel degree contrast to this frame, the foundation of worker thread is in order further to differentiate under the prerequisite not affecting host process real-time, and assisted process is only processed the reference frame therewith time until framing of sending into after the contrast of each host process similarity.

2) channel separation is accurately extracted:

Worker thread is processed as follows:

For each width reference frame image and the two field picture undetermined of input, utilize cvSplit function in opencv by the separation of images of rgb format image bImg, gImg, the rImg for B, G, tri-passages of R, then respectively the similarity of three Color Channels is contrasted, if three passage similarities are all in threshold value, be classified as a class with reference class, otherwise be classified as next class.

S in formula _ir, S _ig, S _ibrepresent respectively R, G, tri-similarity α corresponding to passage of B _r, α _g, α _brepresent respectively the similarity threshold of three path settings.

For the picture frame V that is classified as a class ₁～V _i, get the mean value of all frames using with the frame of mean value difference minimum as key frame, delete all the other frames, to reduce image redundancy.

$\stackrel{\&OverBar;}{V}=\underset{j=1}{\overset{i}{\mathrm{\&Sigma;}}}{V}_j/i---\left(28\right)$

Describedly choose one section of monitor video of 58 minutes and test, test result is as shown in table 1:

Table 1

(6) Moving Objects is taken off and synthesizing of making a summary:

Moving Objects is carried out to foreground detection and mark above, it is main next that but the object at this moment detecting is also not enough to cut apart from background as the object of taking off, for reducing calculated amount, taking rectangle frame that object detected as region, in region, adopt GraphCut algorithm to carry out taking off of Moving Objects, the object of taking off is added to timestamp information deposits Moving Objects database in.

Take out the background image of the different time sections of extracting above and copy, the Moving Objects of taking off being put in chronological order successively to the correspondence position of background image, finally forming the form of video frequency abstract, concrete steps are as follows:

1) take out the background image frame obtaining in first time period;

2) all Moving Objects of taking off in this time period are placed and numbered at the correspondence position of background image in chronological order;

3) object part overlapped in a two field picture is carried out to translucentization processing, in order to avoid affect visual effect;

4) in the time that being greater than certain threshold value, the object number of placing in a two field picture stops placing;

5) copy background image frame and continue placing objects until discharge;

6) take out the background image frame in the next time period and the object of taking off in this time period and place, method is the same, until discharge;

7) all picture frames after synthetic are combined, form final video frequency abstract.

Described in above, be specific embodiments of the invention and the know-why used, if the change of doing according to conception of the present invention, when its function producing does not exceed spiritual that instructions and accompanying drawing contain yet, must belong to protection scope of the present invention.

Claims (5)

1. the monitor video abstraction generating method combining based on object and key frame, is characterized in that comprising the following steps:

1) adjacent two two field pictures are contrasted and obtain changing value, changing value and threshold value compare, and the image that reservation changing value is greater than threshold value is motion fragment, and motion fragment forms new video;

2) new video is evenly divided into several fragments according to video time, utilizes mixture Gaussian background model to the picture frame production background model in each fragment, background model adds that timestamp information deposits in background database;

3) any t moment, current frame image I _twith background model B _trelatively obtain the difference value at (x, y) some place, difference value and threshold value T comparison, the part that difference value is greater than T is Moving Objects, all the other are background;

4) based on Kalman filtering algorithm, Moving Objects is followed the tracks of, the trace information of all Moving Objects deposits in object database;

5) the same Moving Objects detecting is marked, utilize the color characteristic of object to carry out similarity contrast, sort out for the picture frame that similarity is high, to of a sort picture frame calculating mean value, choosing with the frame of mean value difference minimum is key frame;

6) adopt GraphCut algorithm to carry out Moving Objects to take off to marking the region of Moving Objects, from background database, take out the background image of different time sections and copy, the Moving Objects of taking off is put in chronological order successively to the correspondence position of background image, formed video frequency abstract.

2. a kind of monitor video abstraction generating method combining based on object and key frame according to claim 1, is characterized in that described step 5) in, taking the first frame as reference frame, its color histogram is V ₁, the color histogram of subsequent frame is V _i, V _iwith V ₁similarity than for S _i, every frame is divided into MxN piece, and computing formula is:

$S_i=\underset{j=1}{\overset{M\&times;N}{\mathrm{\&Sigma;}}}{S}_{\mathrm{ij}}{M}_j/M\&times;N$

S _ij＝|V _ij-V _1j|

Wherein, S _ijthe similarity value of i frame j piece, M _jit is the weight factor of j piece;

According to above-mentioned formula, picture frame is sorted out, if S _ibe less than or equal to threshold alpha, picture frame and V ₁be classified as a class picture frame L ₁if, α <S _i≤ α ₀, picture frame is classified as and treats framing, if α is >S ₀, picture frame is classified as Equations of The Second Kind L ₂; Then with Equations of The Second Kind L ₂the first frame as reference frame, carry out similarity contrast with each frame below, complete the contrast of all picture frames.

3. a kind of monitor video abstraction generating method combining based on object and key frame according to claim 2, is characterized in that, the described framing for the treatment of is set up to worker thread, adopts the color similarity contrast of channel separation, specifically comprises the following steps:

To input reference frame image and two field picture undetermined utilize cvSplit function by the separation of images of rgb format the image for B, G, tri-passages of R, bImg, gImg, rImg, three of reference frame image and two field picture undetermined Color Channels are carried out respectively to similarity contrast, if the similarity of three passages all in threshold value two field picture undetermined and reference frame image be same class, otherwise treat that framing is classified as next class.

4. a kind of monitor video abstraction generating method combining based on object and key frame according to claim 1, it is characterized in that, described same Moving Objects adopts minimum rectangle frame to mark, and rectangle frame is that similarity comparison domain and Moving Objects are taken off region.

5. a kind of monitor video abstraction generating method combining based on object and key frame according to claim 1, is characterized in that described step 6) specifically comprise the following steps:

6.1) take out the background image frame obtaining in first time period;

6.2) all Moving Objects of taking off in this time period are placed and numbered at the correspondence position of background image in chronological order;

6.3) object part overlapped in a two field picture is carried out to translucentization processing;

6.4) in the time that being greater than certain threshold value, the object number of placing in a two field picture stops placing;

6.5) copy background image frame and continue placing objects until discharge;

6.6) take out the background image frame in the next time period and the object of taking off in this time period, repeating step 6.2)-6.5);

6.7) all picture frames after synthetic are combined, form final video frequency abstract.