CN104580830A - Quasi static image anti-jitter method oriented to video monitoring - Google Patents

Abstract

The invention provides a quasi static image anti-jitter method oriented to video monitoring. The method comprises the steps of updating a reference frame periodically, calculating the projection feature vectors in the vertical direction and the horizontal direction of each frame of image, solving the best offset of each frame of image relative to the reference frame, and correcting the current image according to the best offset. The quasi static image anti-jitter method has the advantages that the processing speed is high, and the quasi static scene image anti-jitter effect on video monitoring application is good.

Description

A kind of quasistatic Image Anti dither method of facing video monitoring

Technical field

The present invention relates to technical field of image processing, specifically a kind of quasistatic Image Anti dither method of facing video monitoring.

Background technology

In video surveillance applications, have a large portion video camera to be erected at field, installation environment is relatively more severe, and strong wind weather or earth shock can cause video camera randomized jitter up and down.If when the scene of monitoring is furthered by high power camera lens, the shake in whole face clearly, will have a strong impact on monitoring effect.Video camera is fixed on more stable support and can reduces shake, but this measure is infeasible in a lot of occasion, therefore need the method adopting image procossing, eliminate the shake of picture, improve video effect.

The essence visually producing shake is: in two width adjacent images, identical impact point, and screen is successively presented at different positions.In image processing field, can think there is an affine transformation relationship between this two width image.In theory, as long as obtain this affine transformation relationship, all pixels of a rear width figure are corrected, just can eliminate jitter phenomenon.The method of conventional feature based Point matching asks for affine transformation relationship at present.First the key point (such as angle point etc.) of two width images is extracted respectively, then by calculating the similarity determination mapping relations between key point, thus calculate geometric transform relation (being generally affine transformation relationship), then according to this relation, again by all pixel transform of the second width image to reposition, thus eliminate shake.

Said method has versatility in theory, therefore can process the problem of the various images match in practical application, but amount of calculation is larger, realizes cost higher.In video surveillance applications, because camera views can only produce translational motion up and down, this is a kind of affine transformation of simplification in fact, and therefore we can solve this kind of specific question by new algorithm, thus reduces amount of calculation, reduces and realizes cost.Such trial has the anti-jitter method of video image " in a kind of video monitoring " (Chinese invention patent application CN104079800A), the method carries out piecemeal to image, calculate the textural characteristics of each image block, matching area is searched in image, then obtain translation parameters according to the matching result of multiple image block, jitter correction is carried out to image.The method, relative to the method for Feature Points Matching, decreases amount of calculation.

In video surveillance applications, the monitoring scene most areas being erected at the video camera in field all can not change usually, can think a kind of quasistatic image, and therefore, vertically relative with the projection properties vector of horizontal direction is also stable.

Summary of the invention

The object of the present invention is to provide a kind of quasistatic Image Anti dither method of facing video monitoring, the method asks for translation parameters according to the vertical of image and horizontal direction projection properties vector, and relative to the method for image block coupling, amount of calculation is less.

Technical scheme of the present invention is:

A quasistatic Image Anti dither method for facing video monitoring, comprises the following steps:

(1) input a two field picture, judge whether to there is reference frame, if so, then perform step (5), if not, then perform step (2);

(2) using present frame as reference frame;

(3) calculate the vertical of present frame and horizontal direction projection properties vector, and result of calculation buffering is got up;

(4) present frame is exported;

(5) judge whether to need to upgrade reference frame, if so, then jump to step (2), if not, then perform step (6);

(6) the vertical of present frame and horizontal direction projection properties vector is calculated;

(7) according to present frame and the vertical of reference frame and horizontal direction projection properties vector, the optimized migration amount of present frame is calculated;

(8) present frame is corrected according to optimized migration amount;

(9) result that current frame offset corrects is exported.

The quasistatic Image Anti dither method of described facing video monitoring, in described step (3) and step (6), calculate the vertical of present frame and horizontal direction projection properties vectorial, specifically comprise:

(21) value of pixel each in current frame image is designated as I (i, j), 0≤i < M, 0≤j < N, wherein, represent the abscissa of pixel, j represents the ordinate of pixel, M represents the width of current frame image, and N represents the height of current frame image;

(22) following formulae discovery is adopted to go out the vertical direction projection properties vector of present frame:

$V=\left[\begin{array}{c}v\left(0\right)\\ v\left(1\right)\\ ...\\ v\left(i\right)\\ ...\\ v(M-2)\\ v(M-1)\end{array}\right]$

$\begin{array}{cc}v\left(i\right)=\underset{0\&le;j<N}{\mathrm{\&Sigma;}}I(i,j)& 0\&le;i<M\end{array}$

Wherein, V represents the vertical direction projection properties vector of present frame, and be a M dimensional vector, v (i) represents that i-th of the vertical direction projection properties vector of present frame maintains number;

(23) following formulae discovery is adopted to go out the horizontal direction projection properties vector of present frame:

$H=\left[\begin{array}{c}h\left(0\right)\\ h\left(1\right)\\ ...\\ h\left(j\right)\\ ...\\ h(N-2)\\ h(N-1)\end{array}\right]$

$\begin{array}{cc}h\left(i\right)=\underset{0\&le;i<M}{\mathrm{\&Sigma;}}I(i,j)& 0\&le;j<N\end{array}$

Wherein, H represents the horizontal direction projection properties vector of present frame, and be a N dimensional vector, h (j) represents that the jth of the horizontal direction projection properties vector of present frame maintains number.

The quasistatic Image Anti dither method of described facing video monitoring, in described step (7), according to present frame and the vertical of reference frame and horizontal direction projection properties vector, calculates the optimized migration amount of present frame, specifically comprises:

(31) following formulae discovery is adopted to go out present frame relative to reference frame offset index in the horizontal direction:

$V_{\mathrm{diff}}\left(m\right)=\frac{1}{M-m}\underset{m\&le;i<M}{\mathrm{\&Sigma;}}\left|\right|(v\left(i\right)-v_b(i-m))\left|\right|$

Wherein, V _diffoffset index when () represents that present frame offsets m pixel in the horizontal direction relative to reference frame m, v _b(i-m) represent that the i-th-m of the vertical direction projection properties vector of reference frame maintains number;

(32) following formulae discovery is adopted to go out present frame relative to reference frame offset index in vertical direction:

$H_{\mathrm{diff}}\left(n\right)=\frac{1}{N-n}\underset{n\&le;j<N}{\mathrm{\&Sigma;}}\left|\right|(h\left(j\right)-h_b(j-n))\left|\right|$

Wherein, H _diffoffset index when () represents that present frame offsets n pixel in vertical direction relative to reference frame n, h _b(j-n) represent that the jth-n of the horizontal direction projection properties vector of reference frame maintains number;

(33) obtain and make V _diffm m that () is minimum, namely obtains present frame relative datum frame optimized migration amount m in the horizontal direction _opt;

(34) obtain and make H _diffn n that () is minimum, namely obtains present frame relative to reference frame optimized migration amount n in vertical direction _opt.

The present invention when shake occurs image for the firm not video camera of support, effectively can suppress the jitter amplitude of image, improve subjective effect, and calculates very fast, and practicality is good.

Accompanying drawing explanation

Fig. 1 is method flow diagram of the present invention.

Embodiment

Below, the present invention is further illustrated with specific embodiment by reference to the accompanying drawings.

As shown in Figure 1, a kind of quasistatic Image Anti dither method of facing video monitoring, comprises the following steps:

Step S01, input a two field picture, judge whether to have cushioned reference frame, if so, then perform step S05, if not, then perform step S02;

Step S02, using present frame as reference frame;

Vertical and the horizontal direction projection properties vector of step S03, calculating present frame, and result of calculation buffering is got up;

The acquiring method of projection properties vector is:

The value of pixel each in image is designated as I (i, j), 0≤i < M, 0≤j < N, i is the abscissa of this pixel, and j is the ordinate of this pixel, and M is the width of image, and N is the height of image; Vertical direction projection properties vector is a M dimensional vector, is designated as V, and as shown in formula [1], v (i) represents that i-th of the vertical direction projection properties vector of present frame maintains number, as shown in formula [2]; Horizontal direction projection properties vector is a N dimensional vector, is designated as H, and as shown in formula [3], h (j) represents that of the horizontal direction projection properties vector of present frame maintains number, as shown in formula [4]; After calculating, result of calculation buffering is got up;

Formula [1]:

$V=\left[\begin{array}{c}v\left(0\right)\\ v\left(1\right)\\ ...\\ v\left(i\right)\\ ...\\ v(M-2)\\ v(M-1)\end{array}\right]$

Formula [2]:

$\begin{array}{cc}v\left(i\right)=\underset{0\&le;j<N}{\mathrm{\&Sigma;}}I(i,j)& 0\&le;i<M\end{array}$

Formula [3]:

$H=\left[\begin{array}{c}h\left(0\right)\\ h\left(1\right)\\ ...\\ h\left(j\right)\\ ...\\ h(N-2)\\ h(N-1)\end{array}\right]$

Formula [4]:

$\begin{array}{cc}h\left(i\right)=\underset{0\&le;i<M}{\mathrm{\&Sigma;}}I(i,j)& 0\&le;j<N\end{array}$

Step S04, present frame to be exported;

Step S05, judge whether to need to upgrade reference frame, renewal according to being: compile a sequence number increased progressively to each frame, when sequence number divides exactly certain normal integer T, be just judged as YES, otherwise be judged as NO, T is larger, and renewal is slower, and the less renewal of T is faster; In general, will be adjusted to proper between 1 ~ 100 second the update cycle; If the judgment is Yes, then perform step S02, if the judgment is No, then perform step S06;

Vertical and the horizontal direction projection properties vector of step S06, calculating present frame, computational methods are see step S03;

Step S07, according to present frame and the vertical of reference frame and horizontal direction projection properties vector, calculate the optimized migration amount of present frame, computational methods are:

When present frame offsets m pixel in the horizontal direction relative to reference frame, offset index V _diff(m) computational methods as shown in formula [5], wherein v _b(i-m) represent that the i-th-m of the vertical direction projection properties vector of reference frame maintains number;

When present frame offsets n pixel in vertical direction relative to reference frame, offset index H _diff(n) computational methods as shown in formula [6], wherein h _b(j-n) represent that the jth-n of the horizontal direction projection properties vector of reference frame maintains number;

Formula [5]:

$V_{\mathrm{diff}}\left(m\right)=\frac{1}{M-m}\underset{m\&le;i<M}{\mathrm{\&Sigma;}}\left|\right|(v\left(i\right)-v_b(i-m))\left|\right|$

Formula [6]:

$H_{\mathrm{diff}}\left(n\right)=\frac{1}{N-n}\underset{n\&le;j<N}{\mathrm{\&Sigma;}}\left|\right|(h\left(j\right)-h_b(j-n))\left|\right|$

The maximum offset that note horizontal direction is possible is MaxM, then m interval is [-MaxM, MaxM], obtains and makes V _diffm m that () is minimum is exactly the optimized migration amount of horizontal direction;

The maximum offset that note vertical direction is possible is MaxN, then n interval is [-MaxN, MaxN], obtains and makes H _diffn n that () is minimum is exactly the optimized migration amount of vertical direction;

Step S08, according to optimized migration amount correct present frame;

If step S07 calculates horizontal direction optimized migration, amount is m _opt, then by present frame integral level direction translation m _optindividual pixel, if step S07 calculates vertical direction optimized migration, amount is n _opt, then by present frame overall vertical direction translation n _optindividual pixel;

Step S09, the result corrected by current frame offset export.

The above execution mode is only be described the preferred embodiment of the present invention; not scope of the present invention is limited; under not departing from the present invention and designing the prerequisite of spirit; the various distortion that those of ordinary skill in the art make technical scheme of the present invention and improvement, all should fall in protection range that claims of the present invention determine.

Claims (3)

1. a quasistatic Image Anti dither method for facing video monitoring, is characterized in that, comprise the following steps:

(1) input a two field picture, judge whether to there is reference frame, if so, then perform step (5), if not, then perform step (2);

(2) using present frame as reference frame;

(3) calculate the vertical of present frame and horizontal direction projection properties vector, and result of calculation buffering is got up;

(4) present frame is exported;

(5) judge whether to need to upgrade reference frame, if so, then jump to step (2), if not, then perform step (6);

(6) the vertical of present frame and horizontal direction projection properties vector is calculated;

(7) according to present frame and the vertical of reference frame and horizontal direction projection properties vector, the optimized migration amount of present frame is calculated;

(8) present frame is corrected according to optimized migration amount;

(9) result that current frame offset corrects is exported.

2. the quasistatic Image Anti dither method of facing video monitoring according to claim 1, is characterized in that, in described step (3) and step (6), calculate the vertical of present frame and horizontal direction projection properties vectorial, specifically comprise:

(21) value of pixel each in current frame image is designated as I (i, j), 0≤i < M, 0≤j < N, wherein, i represents the abscissa of pixel, and j represents the ordinate of pixel, M represents the width of current frame image, and N represents the height of current frame image;

(22) following formulae discovery is adopted to go out the vertical direction projection properties vector of present frame:

$V=\left[\begin{array}{c}v\left(0\right)\\ v\left(1\right)\\ ...\\ v\left(i\right)\\ ...\\ v(M-2)\\ v(M-1)\end{array}\right]$

$v\left(i\right)=\underset{0\&le;j<N}{\mathrm{\&Sigma;}}I(i,j),0\&le;i<N$

Wherein, V represents the vertical direction projection properties vector of present frame, and be a M dimensional vector, v (i) represents that i-th of the vertical direction projection properties vector of present frame maintains number;

(23) following formulae discovery is adopted to go out the horizontal direction projection properties vector of present frame:

$H=\left[\begin{array}{c}h\left(0\right)\\ h\left(1\right)\\ ...\\ h\left(j\right)\\ ...\\ h(N-2)\\ h(N-1)\end{array}\right]$

$h\left(j\right)=\underset{0\&le;i<N}{\mathrm{\&Sigma;}}I(i,j),0\&le;j<N$

Wherein, H represents the horizontal direction projection properties vector of present frame, is a N dimensional vector, and h (j) represents that the of the horizontal direction projection properties vector of present frame maintains number.

3. the quasistatic Image Anti dither method of facing video monitoring according to claim 2, it is characterized in that, in described step (7), according to present frame and the vertical of reference frame and horizontal direction projection properties vector, calculate the optimized migration amount of present frame, specifically comprise:

(31) following formulae discovery is adopted to go out present frame relative to reference frame offset index in the horizontal direction:

$V_{\mathrm{diff}}\left(m\right)=\frac{1}{M-m}\underset{m\&le;i<M}{\mathrm{\&Sigma;}}\left|\right|(v\left(i\right)-v_b(i-m))\left|\right|$

Wherein, V _diffoffset index when () represents that present frame offsets m pixel in the horizontal direction relative to reference frame m, v _b(i-m) represent that the i-th-m of the vertical direction projection properties vector of reference frame maintains number;

(32) following formulae discovery is adopted to go out present frame relative to reference frame offset index in vertical direction:

$V_{\mathrm{diff}}\left(n\right)=\frac{1}{N-n}\underset{n\&le;j<N}{\mathrm{\&Sigma;}}\left|\right|(h\left(j\right)-h_b(j-n))\left|\right|$

Wherein, H _diffoffset index when () represents that present frame offsets n pixel in vertical direction relative to reference frame n, h _b(j-n) represent that the jth-n of the horizontal direction projection properties vector of reference frame maintains number;

(33) obtain and make V _diffm m that () is minimum, namely obtains present frame relative to reference frame optimized migration amount m in the horizontal direction _opt:

(34) obtain and make H _diffn n that () is minimum, namely obtains present frame relative to reference frame optimized migration amount n in vertical direction _opt.