CN101840055B - Video auto-focusing system based on embedded media processor - Google Patents

Abstract

The invention relates to a video auto-focusing system based on an embedded media processor. The system comprises a video collection module, an image acquisition and processing module, a focusing search module and a focusing execution module, wherein the acquisition and processing module is used to perform noise processing to the obtained original video image data and extract Y-component in the data, wherein the induction area selection algorithm based on region segmentation is adopted to extract focus area, the Tenengrad function based on Sobel gradient operator is adopted to calculate definition; the focusing search module is used to determine the moving direction of the camera lens and calculate the required step length (mov_step = step*level) of the lens; the position corresponding to the maximum definition value is used as focus position; and the focusing execution module is used to transmit control signals to a stepping motor used for driving the lens so as to adjust the position of the lens, according to the determined moving direction and step length of the lens in the focusing search module. By using the video auto-focusing system, the focusing accuracy and stability can be effectively increased, and the smoothness is improved.

Description

Video auto-focusing system based on embedded media processor

Technical field

The present invention relates to the video auto-focusing technical field, especially a kind of video auto-focusing system.

Background technology

The video auto-focusing technology is one of gordian technique of modern machines vision.When people carry out obtaining of picture rich in detail through video camera, often need the long period to face toward scenery and focus.Owing to focus on the complicacy of environment, in the focusing process, be difficult to guarantee the flatness that focuses on, promptly the readability of image changes inconsistent.Therefore, how to realize that quick, accurate, level and smooth video auto-focusing seems very important.

The video auto-focusing technology that is directed to video camera at present mainly contains the initiatively type of focusing and the passive type of focusing.So because do not need extra servicing unit to be widely used in the digital camera based on the passive type automatic focus of image.The Chinese patent " auto focusing method and use its automatic focusing system " that such as publication number is 101034198 (application number is 200710079748.5) is through being provided with a plurality of active windows of a plurality of peripheral windows around center window and the center window, and the automatic focus value that assigns weight and calculate per step to a plurality of active windows.But image in the focusing process, still can occur by clear-fuzzy phenomenon, it is unsmooth promptly to focus on process.Such as publication number be the Chinese patent " automatic-focusing camera " of 1199869 (application number is 98107054.X) adopted Hi-pass filter extract collection image high-frequency signal and carry out focused search as focusing on evaluation of estimate; But this method is subject to The noise and does not carry out the selection of focal zone, makes focusing stability not high.

Passive type focalizer that these are traditional and method are prone to affected by environment in focusing stationarity and reliability; And the way of selecting for focal zone generally adopts fixing selection strategy such as center-spot or panorama focusing; But therefore the actual often more complicated of scene that focuses on has caused the accuracy and the poor stability that focus on.Simultaneously do not consider that in the focusing process flatness of focusing makes that people's subjective vision effect is relatively poor.The focusing flatness here refers to scene can only become clear from fuzzy in the focusing process, and can not occur bluring-clear-fuzzy this phenomenon that changes repeatedly.

Summary of the invention

The deficiency of, flatness difference low with stability for the accuracy of the focusing that overcomes existing video auto-focusing system, the present invention provides a kind of accuracy that can effectively improve focusing and stability, improves the video auto-focusing system based on embedded media processor of flatness.

The technical solution adopted for the present invention to solve the technical problems is:

A kind of video auto-focusing system based on embedded media processor comprises: video acquisition module, in order to obtain the raw video image data from camera; The image acquisition process module; Be used for the raw video image data of obtaining are carried out noise processed; And extract Y component wherein, and adopt induction zone zone selection algorithm to extract focal zone based on Region Segmentation, adopt Tenengrad function to carry out sharpness computation based on the Sobel gradient operator; The focused search module; Be used for the direction that definite camera lens moves; And according to current zoom value (the zoom value here is a lens parameters; Be used for characterizing the amplification and the minification of camera lens, reflected the focal length length of variable lens) decide the basic step-length step of search to read the definition values of current lens location image, establish its value and be designated as f ₁And the definition values f of previous frame image ₂Compare and calculate the sharpness rate of change: Δ f=(f ₁-f ₂)/f ₂, calculate step-length coefficient value level, the required mobile step-length mov_step=step of camera lens * level according to Δ f; The definition values that compares present frame and previous frame image does not arrive peak region if the definition values increase is then explained, has arrived peak region if sharpness reduces then judgement, and the position of value of maximum articulation correspondence is as focal position; Focus on execution module, be used for the direction and the step-length that move according to the camera lens that the focused search module is confirmed, to the stepper motor that the drives camera lens adjustment lens location that transmits control signal.

As preferred a kind of scheme: in the said image acquisition process module, based on the process of the induction zone zone selection algorithm of Region Segmentation be: the raw video image data of input are divided into 4 * 4 zonules.Calculate each regional definition values w then _i, wherein i=1...16 compares the definition values w in 16 zones _i, find out maximal value w _Max, its corresponding region is designated as (x as seed ₀, y ₀), with (x ₀, y ₀) be the center, consider (x ₀, y ₀) 4 fields zones (x, y), if (x y) satisfies similarity criterion, then will (x is y) with (x ₀, y ₀) merge as similar area, simultaneously with (x y) is pressed into storehouse, from storehouse, takes out a zone, is used as seed (x to it ₀, y ₀), continue above-mentioned combining step, when storehouse was sky, then Region Segmentation finished, and the zone that is merged is the area-of-interest that chooses.

As preferred another kind of scheme: in the said focused search module, level value calculation criterion is as shown in table 1 below:

Table 1

Wherein, Δ f is the sharpness rate of change, and level representes the coefficient of step-length, and dir representes the camera lens moving direction, and times representes the focused condition counting; When Δ f be on the occasion of the time, use minimum step to move promptly level=1 this moment when the sharpness rate of change surpasses 10%; When the sharpness rate of change increases (representing with symbol " ++ " in the table 1) certainly less than 10% level value, wherein make the level value from subtracting (representing with symbol "--" in the table 1) again 5%～10%; When Δ f was negative value, threshold setting was 5%, and promptly the absolute value of rate of change level=1 greater than 5% time is reverse with the camera lens moving direction simultaneously; When the absolute value of rate of change less than 5% the time, adopt level to skip the local extremum zone from the mode that increases, keep the camera lens moving direction constant simultaneously.

As preferred another scheme: in the said focused search module, confirm that the process of the direction that camera lens moves is: continuously camera lens is moved three positions with minimum step, read three two field pictures, its definition values relatively, the definition values of establishing three images of positions is s ₁, s ₂, s ₃, and if only if s ₁＜s ₂＜s ₃The judgement direction is a negative direction, s ₁＞s ₂＞s ₃The judgement direction is a positive dirction, and all the other situation are then explained and do not found direction, then recomputate the moving direction of camera lens.

Further, in said image acquisition process module, adopt parallel mode to carry out IMAQ.

Further, the formula of said sharpness computation is:

$F\left(I\right)=\underset{x}{\mathrm{\Σ}}\underset{y}{\mathrm{\Σ}}\left|G_x(x,y)\right|+\left|G_y(x,y)\right|---\left(2\right)$

In the following formula, G _x(x, y), G _y(x y) is gradient magnitude, and F (I) is exactly the definition values of the image I that calculates;

Wherein, the G of gradient magnitude _x(x, y), G _y(x, calculating y) with in the image I (x y) carries out convolution for the eight neighborhood subgraphs at center and Sobel gradient operator template and obtains, and the convolution formula is:

$\left\{\begin{array}{c}{G}_x(x,y)=I_8(x,y)*S_x\\ G_y(x,y)=I_8(x,y)*S_y\end{array}\right.---\left(3\right)$

Wherein, I ₈(x, y) expression with in the image I (x y) is the eight neighborhood subgraphs at center, S _xAnd S _yBe the Sobel gradient operator, its operator template is:

$S_x=\left[\begin{array}{ccc}-1& 0& 1\\ -2& 0& 2\\ -1& 0& 1\end{array}\right]$ With $S_y=\left[\begin{array}{ccc}1& 2& 1\\ 0& 0& 0\\ -1& -2& -1\end{array}\right]---\left(4\right).$

Technical conceive of the present invention is: adopt low-power consumption, high performance embedded media processor to realize that quick, accurate, level and smooth video focuses on.In order to realize this function, the present invention need solve following technical matters: (1), need a kind of new focused search strategy for the flatness that guarantees image change in the focusing process.(2), in order to increase the accuracy of focusing, the system of selection of a kind of focal zone effectively of needs.(3), in image acquisition process, how to change original serial operation, maximally utilise processor performance and improve focusing speed.

In the focused search module, the present invention proposes the long searching algorithm of adaptivity multistep.Its algorithm principle is: decide the basic step-length step of search earlier according to current zoom value, the promptly minimum step-length that moves.Rate of change according to the definition values of present frame and previous frame decides step-length coefficient level then, has so just obtained the step-length mov_step=step that ultimate demand moves * level.Because the value of step is come respective change according to the zoom value, so it is bigger than normal and cause converging to the situation of focal position minimum focusing amount can not occur.Because step-length can dynamically be adjusted by the image definition rate of change, make moving of whole step-length can both guarantee that image definition changes steadily simultaneously, promptly people's subjective quality is good when video focuses on, thereby has solved the problem of level and smooth focusing on the whole.

In the image acquisition process module, adopted area-of-interest selection algorithm based on Region Segmentation.Focal zone is selected can reduce data processing amount on the one hand, accelerates focusing speed; Through choosing area-of-interest, eliminate of the influence of non-area-of-interest on the other hand, improve the focusing degree of accuracy the evaluation function curve.Its algorithm thought is: at first the image division with input becomes 4 * 4 zonules.Calculate each regional definition values w then _i, i=1...16 wherein.The definition values w that compares 16 zones _i, find out maximal value w _Max, its corresponding region is designated as (x as seed ₀, y ₀).With (x ₀, y ₀) be the center, consider (x ₀, y ₀) 4 fields zones (x, y), if (x y) satisfies similarity criterion, then will (x is y) with (x ₀, y ₀) merge as similar area, simultaneously with (x y) is pressed into storehouse.From storehouse, take out a zone, be used as seed (x to it ₀, y ₀), continue the combining step of front.When storehouse was sky, then Region Segmentation finished, and the zone that is merged is the area-of-interest that chooses.

In the image acquisition process module, the present invention proposes image pipeline and gather notion, improve focusing speed through rational parallel work-flow.Image pipeline is gathered notion: the process of image once collection is divided into four parts, and the A step is for sending order, the position of adjustment lens.The B step is adjusted to assigned address for inquiry up to lens position.The C step drives for configuration collection, starts IMAQ.The D step is that IMAQ finishes, and carries out the automatic focus analysis.Be illustrated in figure 3 as the string type image acquisition process, at first analyze the needed time that once focuses on: t=(t ₁+ t ₂)+t ₃+ t ₄T wherein ₁And t ₂By lens assembly decision, t ₃By two factors decisions of data volume of the moment that starts collection and collection, t ₄By automatic focus analytical algorithm complexity and two factor decisions of processor host frequency.Because video acquisition is to send into buffer queue through PPI, calling through DMA, view data is sent into the SDRAM zone of appointment.If gather a sub-picture to time of internal memory be t _s, then needed image memory unit number n is at least n=t/t _sIn order to save focal time; It is as shown in Figure 4 that the present invention adopts parallel mode to carry out IMAQ; Be that PPI opens video acquisition always, through DMA constantly with video acquisition in the image memory unit of SDRAM, image acquisition step C only need arrive and find the image of corresponding address to get final product in the storage unit like this; Thereby saved the time of IMAQ, the time that once focus on this moment becomes t ¹=(t ₁+ t ₂)+t ₄, the image memory unit number of needed SDRAM is n ¹=t ¹/ t _sThe method of from internal memory, seeking the corresponding address image is: in steps A, write down current internal memory numbering m, 1＜m＜n ¹Then the address number from image storage unit is among the step C:

m ¹＝[(t ₁+t ₂)/t _s+m]％n ¹ 1＜m ¹＜n ¹

The effect that the present invention is useful is: (1), do not need extra servicing unit just can realize quick, accurate, level and smooth video auto-focusing.(2), image change is level and smooth in whole focusing process, people's subjective vision property is good.

Description of drawings

Fig. 1 is the video auto-focusing system schematic diagram.

Fig. 2 is a video auto-focusing system primary controller design drawing.

Fig. 3 is a string type IMAQ process flow diagram.

Fig. 4 is a parallel type IMAQ process flow diagram.

Fig. 5 is the video auto-focusing system main program flow chart.

Fig. 6 is the automatic focus subroutine flow chart.

Fig. 7 is an image definition counting subroutine process flow diagram.

Embodiment

Below in conjunction with accompanying drawing the present invention is further described.

With reference to Fig. 1～Fig. 7, a kind of video auto-focusing system based on embedded media processor comprises: video acquisition module 7, in order to obtain the raw video image data from camera; Image acquisition process module 8; Be used for the raw video image data of obtaining are carried out noise processed; And extract Y component wherein, and adopt induction zone zone selection algorithm to extract focal zone based on Region Segmentation, adopt Tenengrad function to carry out sharpness computation based on the Sobel gradient operator; Focused search module 9 is used for the direction that definite camera lens moves, and decides the basic step-length step of search according to current zoom value; Read the definition values of current lens location image, establish its value and be designated as f ₁And the definition values f of previous frame image ₂Compare and calculate the sharpness rate of change: Δ f=(f ₁-f ₂)/f ₂, calculate step-length coefficient value level, the required mobile step-length mov_step=step of camera lens * level according to Δ f; The definition values that compares present frame and previous frame image does not arrive peak region if the definition values increase is then explained, has arrived peak region if sharpness reduces then judgement, and the position of value of maximum articulation correspondence is as focal position; Focus on execution module 10, be used for the direction and the step-length that move according to the camera lens that the focused search module is confirmed, each stepper motor that drives camera lens sends crowd fighting and adjusts lens location through control signal.

In the present embodiment, comprise lens 2, CCD 3, FPGA 4, SDRAM 5, primary controller DSP 6 based on the video auto-focusing system of embedded media processor.Described DSP 6 comprises video acquisition module 7, image acquisition process module 8, focused search module 9 again and focuses on execution module 10.Described FPGA 4 links to each other with video acquisition module 7 among the DSP 6.Described SDRAM 5 links to each other with DSP 6.Focusing execution module 10 among the described DSP 6 links to each other with the stepper motor of controls lens.Object 1 at first images on the CCD 3 through lens 2, through FPGA 4 the raw video image data that CCD 3 exports is formatd processing then, and output meets the vision signal of BT.656 standard.Video acquisition module 7 among the DSP 6 collects the specify image storage area among the SDRAM 5 with the video standard signal of FPGA4 output.Image acquisition process module 8 gets current lens location from the designated memory cell of SDRAM 5 image at first carries out noise processed to the image that gets access to, and extracts Y component wherein.Then adopt area-of-interest selection algorithm to extract focal zone, use Tenengrad function to carry out sharpness computation at last based on the Sobel gradient operator based on Region Segmentation.Whether focused search module 9 at first judges camera lens at focal position according to the sharpness evaluation of estimate, if then do not calculate required mobile direction of next step camera lens and step-length, otherwise maintenance lens focus invariant position is accomplished in explanation focusing.Focus on execution module 10 and calculate camera lens moving direction and step-length, send camera lens control signal adjustment lens location for the stepper motor that drives camera lens, thereby realize the focusing of camera lens is moved according to focused search module 9.

With reference to shown in Figure 5, be the video auto-focusing system main program flow chart.Its concrete workflow is as follows: at first carry out the initialization of DSP; The DSP initialization need dispose in have: PPL clock configuration; The configuration of EBIU external bus interface, the external interrupt configuration behind FLASH memory configurations, the DMA, sdram memory configuration and PPI data transmit configuration.Then begin to start video acquisition, only need configure PPI and DMA here after, main control chip is sent to CCD camera capture video data the SDRAM image memory region of appointment automatically.Next carry out camera lens control through button, function is divided into: turn left, turn right, go up change, commentaries on classics, zoom adjustment, manual focusing, automatic focusing and quit a program down.After button carries out automatic focusing, then get into the automatic focus program.

With reference to shown in Figure 6, be the automatic focus program flow diagram.At first need judge the direction that camera lens moves, adopt here with minimum step and continuously camera lens is moved three positions, read three two field pictures, relatively its definition values.If the definition values of three images of positions is s ₁, s ₂, s ₃And if only if s ₁＜s ₂＜s ₃The judgement direction is a negative direction, s ₁＞s ₂＞s ₃The judgement direction is a positive dirction, and all the other situation explain then and do not find direction that then program recomputates the moving direction of camera lens.After lens direction is confirmed, then from internal memory, read current lens location image, and call the definition values calculating that the sharpness computation subroutine is carried out present image, establish its value and be designated as f ₁Then and the definition values f of previous frame image ₂Compare and calculate the sharpness rate of change: Δ f=(f ₁-f ₂)/f ₂Calculate step-length level value according to Δ f, it is as shown in table 1 below to provide level value calculation criterion below:

Table 1

Δ f is the sharpness rate of change in the last table, and level representes the coefficient of step-length, and dir representes the camera lens moving direction, and times representes the focused condition counting.It is to decide according to people's in the experiment subjective observation result that the threshold value of sharpness rate of change is divided, and in experiment, finds to be no more than in 10% when the sharpness rate of change, and human eye can not differentiated the sharpness difference of two width of cloth images.So among the present invention Δ f be on the occasion of the time, use minimum step to move promptly level=1 this moment when the sharpness rate of change surpasses 10%; When the sharpness rate of change increases (representing with symbol " ++ " in the table 1) certainly less than 10% level value, wherein make the level value from subtracting (representing with symbol "--" in the table 1) again 5%～10%, prevent to increase too fast.When Δ f is negative value, occur in order to prevent from the clear phenomenon that fogs, be decided to be 5% to threshold value here, promptly the absolute value of rate of change level=1 greater than 5% time is reverse with the camera lens moving direction simultaneously; When the absolute value of rate of change less than 5% the time, explain this moment slippage seldom, and might be noise or local extremum, so adopt level to skip the local extremum zone, keep the camera lens moving direction constant simultaneously from the mode that increases.The basic step-length step that combine the zoom value to calculate this moment, just can calculate camera lens is required mobile step-length mov_step=step * level.Be to focus on Rule of judgment at last, general focal position judges it all is when definition values reduces when, has thought peak value at this moment, then the pairing position of the definition values of maximum as focal position.But the easy like this local extremum that is absorbed in.And adopt repeatedly judgment mechanism among the present invention, just explained when the situation that occurs the peak value minimizing continuously and crossed peak value.Definition global variable times in whole focused search process, its initial value is made as 0.And if only if when Δ f is negative value, and times is the zero clearings of all the other situation from increasing.When the times value equals T (T is that threshold value is decided according to actual conditions), judge that then camera lens is at focal position at this moment like this.

With reference to shown in Figure 7, image definition counting subroutine process flow diagram.At first carry out the noise processed of image, adopt the medium filtering of 5 * 5 templates to suppress disturbing pulse and point-like noise here.Carry out image Y component extraction then, video format is YUV4:2:2, only need the memory copying of Y component be come out to be used to carry out ensuing focal zone here and detect.The algorithm flow of selecting based on the Region Segmentation area-of-interest is:

Step1: image division is become 4 * 4 zonules.

Step2: calculate wherein i=1...16 of each regional definition values

.

Step3: find out maximal value w _Max, its corresponding region is designated as (x as seed ₀, y ₀).

Step4: with (x ₀, y ₀) be the center, consider (x ₀, y ₀) 4 fields zones (x, y), if (x y) satisfies similarity criterion, then will (x is y) with (x ₀, y ₀) merge as similar area, simultaneously with (x y) is pressed into storehouse.

Step5: from storehouse, take out a zone, be used as seed (x to it ₀, y ₀), get back to the Step4 step.

Step6: when storehouse was sky, then Region Segmentation finished, and the zone that is merged is the area-of-interest that chooses.

Next carry out image definition and calculate, adopt improvement Tenengrad function to carry out sharpness computation here based on the Sobel gradient operator.If the image of input is I, based on the Tenengrad function definition of Sobel gradient operator shown in (1).

$F\left(I\right)=\underset{x}{\mathrm{\Σ}}\underset{y}{\mathrm{\Σ}}\sqrt{G_x^2(x,y)+G_y^2(x,y)}---\left(1\right)$

G in the formula (1) _x(x, y), G _y(x y) is gradient magnitude.F (I) is exactly the definition values of the image I that calculates.Though the Tenengrad function calculation precision based on the Sobel gradient algorithm is high, therefore calculated amount is big makes improvements, and the Tenengrad function after the improvement is suc as formula shown in (2).

$F\left(I\right)=\underset{x}{\mathrm{\Σ}}\underset{y}{\mathrm{\Σ}}\left|G_x(x,y)\right|+\left|G_y(x,y)\right|---\left(2\right)$

Tenengrad function after the improvement is by original 2M * N multiplying and M * N extracting operation, and having changed to only needs M * N sub-addition computing (M, N are the length and the width of image I).Gradient magnitude G in the formula _x(x, y), G _y(x, calculating y) is so that (x y) carries out convolution for the eight neighborhood subgraphs at center and Sobel gradient operator template and obtains, and the convolution formula is suc as formula shown in (3) in the image I.

$\left\{\begin{array}{c}{G}_x(x,y)=I_8(x,y)*S_x\\ G_y(x,y)=I_8(x,y)*S_y\end{array}\right.---\left(3\right)$

I in the formula ₈(x, y) expression with in the image I (x y) is the eight neighborhood subgraphs at center, S _xAnd S _yFor its operator template of Sobel gradient operator suc as formula shown in (4).

$S_x=\left[\begin{array}{ccc}-1& 0& 1\\ -2& 0& 2\\ -1& 0& 1\end{array}\right]$ With $S_y=\left[\begin{array}{ccc}1& 2& 1\\ 0& 0& 0\\ -1& -2& -1\end{array}\right]---\left(4\right)$

At last the image definition value that calculates is delivered in the automatic focus program, and withdraws from subroutine.

Although illustrated and described embodiments of the invention, it should be appreciated by those skilled in the art, under the situation that does not deviate from invention spirit and principle, in accompanying claims and equivalent restricted portion thereof, can make various changes to these instances.

Claims (4)

1. video auto-focusing system based on embedded media processor comprises: video acquisition module, in order to obtain the raw video image data from camera; It is characterized in that: said video auto-focusing system also comprises:

The image acquisition process module; Be used for the raw video image data of obtaining are carried out noise processed; And extract Y component wherein, and adopt induction zone zone selection algorithm to extract focal zone based on Region Segmentation, adopt Tenengrad function to carry out sharpness computation based on the Sobel gradient operator; The focused search module is used for the direction that definite camera lens moves, and decides the basic step-length step of search according to current zoom value; Read the definition values of current lens location image, establish its value and be designated as f ₁And the definition values f of previous frame image ₂Compare and calculate the sharpness rate of change: Δ f=(f ₁-f ₂)/f ₂, calculate step-length coefficient value level, the required mobile step-length mov_step=step of camera lens * level according to Δ f; The definition values that compares present frame and previous frame image does not arrive peak region if the definition values increase is then explained, has arrived peak region if sharpness reduces then judgement, and the position of value of maximum articulation correspondence is as focal position;

Focus on execution module, be used for the direction and the step-length that move according to the camera lens that the focused search module is confirmed, to the stepper motor that the drives camera lens adjustment lens location that transmits control signal;

In the said image acquisition process module, based on the process of the induction zone of Region Segmentation zone selection algorithm be: the raw video image data of input are divided into 4 * 4 zonules, calculate each regional definition values w then _i, wherein i=1...16 compares the definition values w in 16 zones _i, find out maximal value w _Max, its corresponding region is designated as (x as seed ₀, y ₀), with (x ₀, y ₀) be the center, consider (x ₀, y ₀) 4 neighborhoods zones (x, y), if (x y) satisfies similarity criterion, then will (x is y) with (x ₀, y ₀) merge as similar area, simultaneously with (x y) is pressed into storehouse, from storehouse, takes out a zone, is used as seed (x to it ₀, y ₀), continue above-mentioned combining step, when storehouse was sky, then Region Segmentation finished, and the zone that is merged is the area-of-interest that chooses;

In the said focused search module, level value calculation criterion is as shown in table 1 below:

Table 1

Wherein, Δ f is the sharpness rate of change, and level representes the coefficient of step-length, and dir representes the camera lens moving direction, and times representes the focused condition counting; When Δ f be on the occasion of the time, use minimum step to move promptly level=1 this moment when the sharpness rate of change surpasses 10%; When the sharpness rate of change increases less than 5% level value certainly, said from adding symbol " ++ " expression, wherein make the level value from subtracting again 5%～10%, said subtracting with symbol "--" certainly represented; When Δ f was negative value, threshold setting was 5%, and promptly the absolute value of rate of change level=1 greater than 5% time is reverse with the camera lens moving direction simultaneously; When the absolute value of rate of change less than 5% the time, adopt level to skip the local extremum zone from the mode that increases, keep the camera lens moving direction constant simultaneously.

2. the video auto-focusing system based on embedded media processor as claimed in claim 1; It is characterized in that: in the said focused search module; The process of confirming the direction that camera lens moves is: continuously camera lens is moved three positions with minimum step; Read three two field pictures, compare its definition values, the definition values of establishing three images of positions is s ₁, s ₂, s ₃, and if only if s ₁＜s ₂＜s ₃The judgement direction is a negative direction, s ₁＞s ₂＞s ₃The judgement direction is a positive dirction, and all the other situation are then explained and do not found direction, then recomputate the moving direction of camera lens.

3. the video auto-focusing system based on embedded media processor as claimed in claim 1 is characterized in that: in said image acquisition process module, adopt parallel mode to carry out IMAQ.

4. the video auto-focusing system based on embedded media processor as claimed in claim 1 is characterized in that: the formula of said sharpness computation is:

$F\left(I\right)=\underset{x}{\mathrm{\Σ}}\underset{y}{\mathrm{\Σ}}\left|G_x(x,y)\right|+\left|G_y(x,y)\right|---\left(2\right)$

In the following formula, G _x(x, y), G _y(x y) is gradient magnitude, and F (I) is exactly the definition values of the image I that calculates;

Wherein, the G of gradient magnitude _x(x, y), G _y(x, calculating y) with in the image I (x y) carries out convolution for the eight neighborhood subgraphs at center and Sobel gradient operator template and obtains, and the convolution formula is:

$\left\{\begin{array}{c}{G}_x(x,y)=I_8(x,y)*S_x\\ G_y(x,y)=I_8(x,y)*S_y\end{array}\right.---\left(3\right)$

Wherein, I ₈(x, y) expression with in the image I (x y) is the eight neighborhood subgraphs at center, S _xAnd S _yBe the Sobel gradient operator, its operator template is:

$S_x=\left[\begin{array}{ccc}-1& 0& 1\\ -2& 0& 2\\ -1& 0& 1\end{array}\right]$ With $S_y=\left[\begin{array}{ccc}1& 2& 1\\ 0& 0& 0\\ -1& -2& -1\end{array}\right]---\left(4\right).$

Images