CN101068342A - Video frequency motion target close-up trace monitoring method based on double-camera head linkage structure - Google Patents

Abstract

A close up tracking and monitoring method of video movement object based on linkage structure of double camera shooting heads includes using a overall view monitor camera shooting head to carry out identification on object in monitoring region to confirm position and direction as well as movement speed of object, sending object information to close up tracking camera shooting head with direction being controlled by rotary table after tracking object is locked out, obtaining more information of locked object by utilizing close up tracking camera shooting head to display close up picture of locked object.

Description

Video frequency motion target feature based on the dual camera linkage structure is traced and monitored method

Technical field

The invention belongs to technical field of video monitoring, relate to a kind of method for supervising of video frequency motion target, be specifically related to a kind of method of utilizing the dual camera linkage structure that video frequency motion target is monitored.

Background technology

Along with the continuous development of computer video image processing techniques, and the pressing for of field such as security protection, anti-terrorism, the intelligent monitoring system requirements is accomplished to a certain degree automation to the behavioural analysis of intrusion target.

In order to guarantee to detect moving target, the camera of video monitoring system is set and mostly is fixed mode greatly.The problem that target detection under the fixed mode exists is: increase the resolution of target if desired, then the scene of Jian Shiing can only concentrate on certain part, for example, and the porch etc. of boarding on airport.Increase monitoring range if desired, then can't guarantee the portrayal of the intrusion target details in the monitoring image.For example, in license board information of doorway, sub-district escape motorcycle etc.

If should monitor to panorama, again suspicious object is judged automatically, and carried out feature and follow the tracks of, just need finish with the pattern of " quiet is moving ".

Summary of the invention

The object of the present invention is to provide a kind of video frequency motion target feature to trace and monitor method, video object is got final product overall view monitoring, can carry out feature again and follow the tracks of based on the dual camera linkage structure.

The technical solution adopted in the present invention is, video frequency motion target feature based on the dual camera linkage structure is traced and monitored method, by an overall view monitoring camera target of monitor area is discerned, determine the position of target, the speed of service and direction, after the locking tracking target, target information is passed to the feature that is turned to by cradle head control follow the tracks of camera, follow the tracks of camera by feature lock onto target is carried out feature, amplifying the back follows the tracks of, the feature picture of display-object, thereby obtain the target more information, this method is carried out according to the following steps

A. the feature that adopts an overall view monitoring camera and to be arranged on the The Cloud Terrace is followed the tracks of camera, adopt the interlock between a computer realization overall view monitoring camera and the feature tracking camera, promptly set up each some position corresponding relation in the captured video pictures of two cameras respectively in the scene picture, according to the position of target in the panoramic shooting head determine the feature camera towards, enable the target of following the tracks of over against needs;

B. utilizing the overall view monitoring camera that the target of monitor area is carried out behavior detects, promptly use the method for image and Video processing that moving target is separated with the background difference of scene, obtain the information of number, the speed of travel, direction of travel and the position of intrusion target;

C. to the above-mentioned detected Target Setting direction of motion, target location, movement velocity, four characteristic parameters of color of object, and note, the method of utilization template matches is followed the tracks of moving target between the frame of video and frame, obtains the motion track information of each target;

D. overall view monitoring camera interaction relation that the target trajectory information that obtains is set up by step a is transferred to feature and follows the tracks of camera, rotation by the control The Cloud Terrace drives the rotation of feature tracking camera, carry out target following, the target with locking is presented in the monitored picture after amplifying real-time and accurately;

E. control feature and follow the tracks of the convergent-divergent of the camera focal length target that furthers, the above-mentioned target in the monitored picture of being locked in that obtains is carried out feature and taken.

The inventive method adopts the dual camera linkage structure, realizes the moving target in the intrusive monitoring visual field is carried out the feature tracking.What the overall view monitoring camera was finished is that a visual angle is wider, and certain regional panorama is monitored.Though the resolution of each target in the panorama is all smaller, be difficult to identification in detail, computer can obtain the running orbit of intrusion target in monitor area, process of action or the like to the automatic analysis of the panorama monitoring video information of acquisition.For this reason, the present invention utilizes the overall view monitoring camera to finish the target that enters monitor area is discerned, and determines position, the speed of service and the direction of target.The overall view monitoring camera is passed to feature tracking camera with the movement velocity and the directional information of detected suspicious object through serial communication after having locked tracking target.Feature is followed the tracks of camera and is responsible for that suspicious object is carried out feature and amplifies the back and follow the tracks of, the feature picture of display-object, thus can obtain the more information of suspicious object, the less or smudgy beyond all recognition deficiency of suspicious object that provides when resulting video has been provided.

Description of drawings

Fig. 1 is that dual camera angle and the irradiation position that the inventive method adopts concerns schematic diagram;

Fig. 2 camera angle and cradle head control angle concern schematic diagram;

The pixel value change curve of certain pixel in Fig. 3 frame of video, abscissa is a time shaft, unit is a frame number; Ordinate is certain normalization pixel value of this point constantly;

The variation track of six kinds of basic exercises of Fig. 4 The Cloud Terrace;

Fig. 5 camera focal length sample fitting result chart, abscissa is the command interval time, ordinate is a multiplication factor.

Embodiment

The present invention is described in detail below in conjunction with the drawings and specific embodiments.

The video frequency motion target feature that the present invention is based on the dual camera linkage structure is traced and monitored method, employing be the structure of two cameras interlock.So-called dual camera linkage structure is meant a fixed overall view monitoring camera is set, and the panorama of finishing whole monitor area monitors.Another is erected at the camera on the rotatable The Cloud Terrace of two degrees of freedom, finishes the feature of lock onto target is followed the tracks of.After the panorama monitoring camera is found target, the speed and the direction of its motion are judged, and driving feature tracking camera rotates tracking along with the motion of target.So-called feature is followed the tracks of, and is meant that the monitoring image of tracking target camera has only target, like this, can make the correct identification of object become possibility because of the resolution that increases target.

Two cameras are by a computer realization interlock.The panoramic shooting head is fixed on certain position, and camera is connected on the video card on the computer, and in computer, computer is handled the vision signal of input with video signal transmission, and judges and when have suspicious object to occur.In case having judged suspicious object occurs, just calculate its position, with and movement velocity and direction, and these parameters are transferred out by motion control card, control the camera on another rotary platform that is erected at two-freedom, turn to the position at target place, camera is adjusted focal length afterwards, makes this camera amplify feature to target and follows the tracks of.After this feature is followed the tracks of camera acquisition target, also vision signal is transferred in the computer by video card, computer is controlled this feature tracking camera and is moved along with the motion of target according to the movement velocity and the direction of target, finishes the feature of target is followed the tracks of.

The dual camera interlock method

After the overall view monitoring camera has locked intrusion target, making another feature follow the tracks of camera can find this object and it is amplified tracking, then need to set up two interaction relations between the camera, promptly set up each some position corresponding relation in the captured video pictures of two cameras respectively in the scene picture.Promptly according to the position of target in the panoramic shooting head determine the feature camera towards, enable the target of following the tracks of over against needs.Introduce the process that this interaction relation is set up below.

As shown in Figure 1, suppose that the imaging point of two cameras overlaps (as the point of the O among Fig. 1), according to the image-forming principle of camera as can be known, the position that the feature camera is aimed at is relevant with respect to the angle of panoramic shooting head with it.

If the direction vector of panorama monitoring camera is Φ, be a fixed value, establish the vectorial φ of being of inceptive direction that feature is followed the tracks of camera, the angle of the two is θ ^{(Φ, φ)}Direction vector with the panoramic shooting head is that benchmark is set up horizontal plane X and vertical projection face Y.On projection of angle to two face, be decomposed into the directions X angle theta _X ^{(Φ, φ)}With Y angular separation θ _Y ^{(Φ, φ)}Here, in the feature camera aligning panoramic shooting head (x, y) point, following as we know from the figure relation:

${\mathrm{\θ}}_X^{(\mathrm{\Φ},\mathrm{\φ})}=k_{x}x---\left(1\right)$

${\mathrm{\θ}}_Y^{(\mathrm{\Φ},\mathrm{\φ})}=k_{y}y---\left(2\right)$

Position coordinates and camera angle are linear.

Such relation has been arranged, when needs allow the feature camera point in the panoramic shooting head certain some time, just can realize by the angle of controlling two cameras.In other words, allow the feature camera point to that (x, y) point are θ with regard to controlling the angle that camera makes it with the panoramic shooting head if desired ^{(Ф, φ)}

This control procedure is that the rotation by The Cloud Terrace realizes.Be exactly to be control to the angle control transformation of two cameras below to two angles of The Cloud Terrace.As shown in Figure 2, the direction vector of panoramic shooting head is Φ, and the angle of it and horizontal plane is made as β, is a fixed value, and the direction vector of establishing the feature camera is φ.Direction of illumination with the panoramic shooting head is a benchmark, sets up X perspective plane and Y perspective plane, corresponds respectively to OAX plane and OAY plane among the figure.

Feature is followed the tracks of camera direction vector φ projection on two planes, do vertical line to two perspective planes respectively from last 1 B of vectorial φ, the vertical point that obtains the X face is B ', and the vertical point of Y face is Y, and Y ' is the subpoint of B on vectorial Φ.So just obtain two camera directions X angle theta respectively _X ^{(Ф, φ)}With Y angular separation θ _Y ^{(Ф, φ)}, corresponding to ∠ AOB ' among Fig. 2, ∠ AOY.On the other hand, according to the structure of The Cloud Terrace as can be known, the rotation angle of The Cloud Terrace horizontal direction is corresponding to α among the figure _X, the rotation angle of vertical direction is corresponding to the α among the figure _Y

Next (x is y) with (α will to set up coordinate exactly _X, α _Y) between relation.That at first, calculates The Cloud Terrace horizontally rotates angle α _XAccording to concerning among the figure, as can be known:

$\tan {\mathrm{\θ}}_X^{(\mathrm{\Φ},\mathrm{\φ})}=\frac{\left|Y^{\′}{B}^{\′}\right|}{\left|Y^{\′}O\right|}---\left(3\right)$

$\tan {\mathrm{\α}}_X=\frac{\left|\mathrm{YB}\right|}{\left|Y{O}^{\′}\right|}---\left(4\right)$

According to projection relation as can be known: | YB|=|Y ' B ' | then can get by formula (3) (4):

$\tan {\mathrm{\α}}_X=\frac{\left|Y^{\′}O\right|}{\left|Y{O}^{\′}\right|}\tan {\mathrm{\θ}}_X^{(\mathrm{\Φ},\mathrm{\φ})}---\left(5\right)$

Among Fig. 2

Then have:

$\left|\mathrm{YO}\right|=\frac{\left|Y^{\′}O\right|}{\cos {\mathrm{\θ}}_Y^{(\mathrm{\Φ},\mathrm{\φ})}}---\left(7\right)$

Can get thus:

Formula (1) (2) formula substitution (10) is obtained α _XFinal computing formula:

Next calculate the vertical angle of rotation α of The Cloud Terrace again _YAccording to concerning among the figure, as can be known:

|BY|＝|BO′|sinα _X (12)

|YO′|＝|BO′|cosα _X (13)

Formula (13) substitution (6) can be got:

Following relation is arranged in right-angled triangle ⊥ BYO:

Following relation is arranged in right-angled triangle ⊥ BO ' O:

With formula (1), (2) substitution (16) obtains α again _YFinal computing formula:

Interaction relation according to the dual camera that obtains above, at first calculate the position of lock onto target in feature tracking camera in the panorama monitoring camera, feature is followed the tracks of camera turn to corresponding position, the object of locking is followed the tracks of in the supervision visual field of camera at feature, afterwards, feature is followed the tracks of camera and is rotated accordingly according to target travel direction and the speed judged, finishes the feature of target is followed the tracks of.After the feature of having finished a target was followed the tracks of, this camera resetted, and waits for the appearance of next target.

The information processing of overall view monitoring camera

The panorama monitoring camera is fixed, the effect of this camera is, the moving target that occurs in the fixing supervision scene is detected, and its behavior carried out simple analysis, obtain number, the speed of travel of intrusion target, the direction of travel of intrusion target, and information such as position, can discern automatically and in time report to the police the suspicious actions of moving target.

Can be provided with voluntarily according to actual needs the warning mode as: invasion warning region, warning line also can be whole monitor areas, follow the tracks of to enter the sequencing that monitors the visual field.After sign was finished, system just detected the target that enters in the warning region, after detecting target, also can add up monitoring the number in the scene, and can the personage in monitoring the visual field be assembled a crowd to judge automatically and report to the police.

1) motion target detection

Describe as moving target with the personage below.So-called moving object detection just is meant that the means of utilization image and Video processing are distinguished the target (people) of motion mutually with the background of scene and separates.The present invention has adopted the method for background being carried out modeling, obtain a background frame (referring to the supervision scene of the driftlessness thing of generation automatically), constantly this background is upgraded afterwards, to guarantee in target detection, can adapt to outdoor different time, the change of background that the variation of the photoenvironment under the different weather causes.

The present invention has adopted the method for sorting out based on pixel grey scale to realize background is carried out modeling, after obtaining not comprise the background frames of moving target, adopts simple frame-to-frame differences method, just can detect moving target.

The pixel grey scale subsumption algorithm is to be based upon hypothesis background pixel gray scale to appear under the prerequisite in the image sequence with maximum probability.This hypothesis is feasible in actual applications, is changeless because only have powerful connections, and moving target is to have blocked background in the short time, and background is not blocked in the most of the time.So the present invention utilizes gray scale difference that respective pixel point gray scale is sorted out, select the background pixel value of the highest gray value of the frequency of occurrences as this point.Concrete algorithm is as follows.

At first, import one section video that includes the N two field picture, afterwards, the grey value profile of each pixel in this N two field picture is added up, in this patent, gray value is through the result after the normalized.That is, the pixel value of establishing current point be f (i, j), then the pixel value x after the normalization (i j) is:

x(i，j)＝f(i，j)/255 (18)

Here, the 255th, the max pixel value of present 8 bitmaps.

Afterwards, judge that the intensity profile in this N two field picture is added up, setting fluctuation range is 0.1, then the gray value after the normalization is divided into 10 value classes, to the array N of given one 10 dimension of each pixel value _{X (i, j)}(k), k=1,2 ..., 10, then have:

N _{X (i, j)}(k)=in the N two field picture, and gray value drop on (k-0.05, k+0.05) number of pixels in the scope }

Statistics is found out N after finishing _{X (i, j)}(k), k=1,2 ..., 10 when being maximum, corresponding k.

That is: N _Max=max{N _{X (i, j)}(k), k=1,2 ..., 10}

Afterwards, calculate its average and variance, obtain the confidential interval of candidate background dot intensity profile.Like this, in fact obtained the background model of this monitoring image.

As shown in Figure 3, be one section change curve after the pixel value normalization with a certain pixel in the 155 frame videos, transverse axis express time axle, unit are frame number.The longitudinal axis is through the pixel value after the normalization.From this curve as can be known, in this video-frequency band, this point can be blocked by object at some constantly, the more part of statistics occurrence frequency is near 0.55 after the normalization, it to be worth as a setting, passes through clustering processing, obtain its distribution average and variance, according to the definition of the confidential interval of statistical mathematics, be background dot with dropping on that point in the background confidential interval is used as, otherwise just be judged as the candidate impact point.

Afterwards, the point that is judged as the candidate target is carried out the specificity analysis of connected domain again, will be judged as the point deletion of non-target, what stay is exactly the point that is judged as target.

To the video pictures of a new frame of input, judge whether corresponding gray values of pixel points has dropped in the confidential interval.If not, this point is the pixel of target, at this moment, background model is constant.If, then show this point for not by the background dot of target occlusion, then according to following formula computation of mean values and variance, revise background model.

Average is upgraded:

x _t+1(i，j)＝( x _t(i，j)+x _t+1(i，j))/2 (19)

Wherein, x _T+1(i, j) the pixel average of expression current time, x _t(i j) is the pixel average of previous moment, x _T+1(i j) is the pixel value of current time.

Obviously,, can realize the memory that fades, make modeling to background, reflect the photoenvironment of current time all the time according to formula (19).

Variance is upgraded:

${{\mathrm{\σ}}^2}_{t+1}(i,j)=\underset{k=1}{\overset{N_{\max }}{\mathrm{\Σ}}}{(x(i,j)-{\stackrel{\‾}{x}}_{t+1}(i,j))}^2---\left(20\right)$

At this moment, be [x as the background frames image that detects moving target _T+1(i, j)] _M*n

System when carrying out monitor task, to the pixel value x of the present frame of input (i, j), adopt the method for frame-to-frame differences to calculate according to following formula:

e(i，j)＝|x(i，j)- x _t+1(i，j)| (21)

If frame-to-frame differences e (i j) greater than prior preset threshold, shows that then this point is an impact point, otherwise, be indicated as background dot, it is used formula (19), (20) are upgraded.

2) tracking of moving target

Computer detects according to top method the vision signal that the overall view monitoring camera photographs, in case found after the target, will write down the characterisitic parameter of detected target.Consider in monitor area, a plurality of targets might appear, and the movement locus of target in the zone constantly changes, in order to prevent error tracking, to a detected Target Setting " direction of motion ", " target location ", " movement velocity ", " color of object " four characteristic parameters.These four characteristic parameters are noted, the method of utilization template matches is followed the tracks of moving target between the frame of video and frame, and write down the movement locus of each target, movement locus by target with and current location drive feature and follow the tracks of camera, allow feature follow the tracks of camera and find the target of locking, and carry out feature and follow the tracks of.

Camera has the regular hour hysteresis because data are sent to the feature tracking, the relation before and after therefore also needing to consider between the frame of video.

Enter the supervision visual field as a plurality of moving targets, if do not intersect between the moving target, native system is distinguished target by the distance between the detected target on the judgement two continuous frames, and target is numbered.Calculate the movement velocity and the direction of each target simultaneously.

Under the situation that occurs intersecting at the movement locus between a plurality of targets, then distinguish target, thereby obtain the movement locus of each target and it is carried out record by target speed, color characteristic, direction character are compared.

To the statistics of velocity to moving target, employing be that the method for the pixel count that moves in different frame of estimating target central point realizes.The analysis of color of object feature then is by people's build and people's daily habits are divided into three parts to color characteristic promptly: head, and upper body, the lower part of the body is through an a certain proportion of average cutting apart and count three Color Channels of each part.To the estimation of target travel direction, mainly be to ask difference to obtain the direction of motion of target when the variation of former frame internal coordinates by target.Distinguish target that we detect according to these three features then and it is numbered, obtain the motion track information of each moving target in real time.

3) warning of suspicious actions

Detection and the method for analyzing automatically to movement objective orbit information have been arranged, just can be implemented under the warning mode of user's setting, target is entered warning region, target pass through the identification of suspicious actions of overall importance such as assemble a crowd in the hot spot region of warning line, target, and after identifying suspicious actions, report to the police.

Feature is followed the tracks of the control of camera

After the overall view monitoring camera has obtained intrusion target,, require feature to follow the tracks of camera and finish " feature shooting " and reach " target following " two actions by the interlock mode of the dual camera introduced previously.For this reason, feature is followed the tracks of the control of camera, and the The Cloud Terrace rotation that in fact needs to finish setting up camera is controlled, and the camera focal length is controlled.

" feature shooting " is exactly the clear demonstration that realizes target by the size of control camera focal length, so that follow-up people's face three-dimensional reconstruction, target identities identification.

" target following " is exactly according to the target signature parameter that transmits, and the control The Cloud Terrace drives camera and rotates, and the target with locking is presented in the monitored picture after amplifying real-time and accurately.

1) cradle head control

Cradle head control is in order to finish the task of " target following ".Can adopt one to be driven by two high accuracy stepping motors on level and the vertical direction, two motors send commands for controlling by the MPC07 motion control card.The MPC07 control card is based on the upper control unit of the stepping motor of PC pci bus, and it and PC constitute the master slave control structure.The work of the management of PC director machine interactive interface and the aspects such as real-time monitoring of control system (for example monitoring of the transmission of the demonstration of the management of keyboard and mouse, system mode, control command, external signal or the like).The MPC07 card is finished all details (comprising the processing, initial point of output, the automatic lifting speed of pulse and direction signal and detection of signal such as spacing or the like) of motion control.

The motion control function of MPC07 control card depends primarily on the movement function storehouse.The movement function storehouse provides many movement functions for the stepping or the SERVO CONTROL of single shaft and multiaxis: single shaft motion, multiaxis self-movement, multi-axis interpolation motion or the like.In addition, for the exploitation of routing motion control system, also provide the gap compensate function.Simply introduce the function and the motion mode of these function correspondences below.

This control card provides six kinds of basic exercise types, is listed in the table 1.Fig. 4 has provided under this control card control, six kinds of basic exercise tracks of The Cloud Terrace.

Six kinds of basic exercise types of table 1MPC07 control card

Function	Function
		con_pmove	With the mobile distance to a declared goal of normal speed (Fig. 4 (a))
fast_pmove	Move distance to a declared goal (Fig. 4 (b)) with trapezoidal speed
		con_vmove	Normal fast continuous motion (Fig. 4 (c)) with appointment
fast_vmove	Remain on after the acceleration and specify continuous motion (Fig. 4 (d)) at a high speed
		con_hmove	Move to initial point (Fig. 4 (e)) with Chang Su
fast_hmove	Move to origin position (Fig. 4 (f)) after the acceleration fast

The movement function that has lifting/lowering speed control system is referred to as (fast) movement function fast, for example: fast_pmove, fast_vmove and fast_hmove.Normal fast movement function then is referred to as Chang Su (con) movement function, as con_pmove, and con_vmove, con_hmove.This card also provides multiple motion modes such as multiaxis self-movement, multi-axis interpolation motion in addition.

On tracking Control, native system adopts this card to do hardware supports, under the VC development environment, after system detects the intended target position, the pulse signal of positional information and control card is through certain transformational relation, control MPC07 sends the pulse signal of appointment, thereby drive feature tracking cam movement is finished the tracking to intended target.

2) camera focus controlling

The focal length that feature is followed the tracks of camera can be provided with in advance according to actual scene, to guarantee in the monitor area scope, to the clear view of people's object detail.

The camera focus controlling is in order to finish target " feature shooting " task.For can clearer record object, for the positive face of the personage of back detects and facial three-dimensional reconstruction lays the first stone, must carry out the shooting of feature, this just must regulate focal length of feature tracking camera.Native system adopts the convergent-divergent of the serial port communication technology control camera focal length target that furthers to take for it with feature.Under the VC development environment, adopt the MSCOMM control to realize serial communication, the control command of focal length has set form, and the time length that sends signal has directly determined the size of focal length variations.

In the table 2, listed the focus controlling command list.

Table 2 camera focus controlling command list

Operation	The serial ports instruction
		The focal length of camera scaling stops	EE0041h
Focal length of camera amplifies beginning	EE004Ch
		Focal length of camera dwindles beginning	EE004Bh

Because the camera that system adopts can't directly obtain the size parameter of focal length, therefore, the present invention has adopted round-about way to obtain the size of focal length.Here, choose the sign of the ratio of the area of target and normal video (being assumed to be 320 * 240 pixels) size as reflection focal length size.In fact the two ratio is not the accurate reflection of real focal length, even because the size of target person is not only relevant with focal length also relevant with other objective factor under identical focal length, shared video ratio under the same focal length is different such as adult and child, but for feature shows, purpose is that target is amplified (or dwindling) as long as can identification just reach requirement to certain proportion, do not need accurately to reflect focal length, so so indirect reflection relation satisfies system requirements fully.

Introduce among the present invention the Mathematical Modeling of camera below again and set up process.This first important feature of making a video recording is that in amplification process, initial size and the ratio that stops size are corresponding with amplifying the blanking time of instructing.So just can carry out modeling to video camera according to these characteristics.

If the initial multiplication factor of video camera is α ₀, T _nVideo camera multiplication factor constantly is α _n

ΔT＝T _k-T _k-1 k∈n (22)

Suppose that Δ T keeps constant and enough little, then can be similar to and think that initial focal length size is linear with the ratio and the Δ T that stop the focal length size in the Δ T time, then have:

$\frac{{\mathrm{\α}}_k}{{\mathrm{\α}}_{k-1}}=\mathrm{A\ΔT}---\left(23\right)$

Wherein, α _kExpression T _kFocal length of camera size constantly, A is a constant.

The rest may be inferred then has:

$\frac{{\mathrm{\α}}_n}{{\mathrm{\α}}_0}=\frac{{\mathrm{\α}}_n}{{\mathrm{\α}}_{n-1}}\·\frac{{\mathrm{\α}}_{n-1}}{{\mathrm{\α}}_{n-2}}\mathrm{\Λ}\frac{{\mathrm{\α}}_1}{{\mathrm{\α}}_0}={\left(\mathrm{A\ΔT}\right)}^n---\left(24\right)$

T blanking time of scaling instruction can be represented as:

t＝nΔT (25)

$n=\frac{t}{\mathrm{\ΔT}}---\left(26\right)$

(26) substitution (24) is got:

$\mathrm{\Δ\α}\left(t\right)=\frac{{\mathrm{\α}}_t}{{\mathrm{\α}}_0}={\left(\mathrm{A\ΔT}\right)}^{\frac{t}{\mathrm{\ΔT}}}---\left(27\right)$

Because A, Δ T is a constant, and then (27) can be converted into:

$\mathrm{\Δ\α}\left(t\right)=\frac{{\mathrm{\α}}_t}{{\mathrm{\α}}_0}=T^t---\left(28\right)$

In the following formula $T={\left(\mathrm{A\ΔT}\right)}^{\frac{1}{\mathrm{\ΔT}}}$ Be constant.

On the other hand, when considering system's scaling focal length, be that the conducting by the decoder repeat circuit realizes with disconnecting, so can have mechanical response time τ during the focal length scaling, then (28) are adjusted:

$\mathrm{\Δ\α}\left(t\right)=\frac{{\mathrm{\α}}_t}{{\mathrm{\α}}_0}=T^{t-\mathrm{\τ}}---\left(29\right)$

The Mathematical Modeling of focal length of camera amplification just has been established like this.

In like manner:, then can obtain the Mathematical Modeling that focal length of camera dwindles because the amplification of focal length is symmetrical with dwindling:

$\mathrm{\Δ\α}\left(t\right)=\frac{{\mathrm{\α}}_0}{{\mathrm{\α}}_t}=T^{t-\mathrm{\τ}}---\left(30\right)$

Next will determine parameter exactly,, the multiplication factor of video camera and command interval time done following sampling test here, experimental result sees Table 3.

Table 3 video camera amplification time t and magnification ratio sampling table

t(s)	Δα	t(s)	Δα	t(s)	Δα
						0.5	1.19	1.4	4.46	2.3	25.00
0.6	1.23	1.5	5.98	2.4	29.64
						0.7	1.49	1.6	5.99	2.5	37.35
0.8	1.78	1.7	7.72	2.6	42.98
						0.9	2.09	1.8	9.68	2.7	52.16
1.0	2.42	1.9	11.11	2.8	64.00
						1.1	3.16	2.0	13.44	2.9	79.01
1.2	3.57	2.1	15.12	3.0	100.00
						1.3	4.01	2.2	19.75	3.1	118.51

According to above sampling test experiments data, above Mathematical Modeling has been carried out match, as shown in figure 11, be that sampled point is carried out the matched curve that obtains after the match.Lines 1 are sample curve, the Mathematical Modeling curve that lines 2 amplify for the focal length of camera that simulates.Like this, just can obtain, the Mathematical Modeling that focal length amplifies is:

$\mathrm{\Δ\α}\left(t\right)=\frac{{\mathrm{\α}}_t}{{\mathrm{\α}}_0}={6.119}^{t-0.5217}---\left(31\right)$

The Mathematical Modeling that focal length dwindles is:

$\mathrm{\Δ\α}\left(t\right)=\frac{{\mathrm{\α}}_0}{{\mathrm{\α}}_t}={6.119}^{t-0.5217}---\left(32\right)$

Control main purpose to focal length of camera is in order to adjust focal length of camera, to make the object of care keep a desirable size in picture, so that observe.According to this requirement, can not need precision is had higher requirement as the The Cloud Terrace Position Control to the control of focal length of camera.Therefore do not need to adopt complicated control method.

Here, according to top definite Mathematical Modeling, following algorithm is adopted in the control of focusing:

The size of object is m in the video pictures, establishes given target expectation size and is M, and then error e is expressed as:

$e=\frac{m}{M}---\left(33\right)$

It is (0.9,1.1) that native system is set the permissible error scope, when e ∈ (0.9,1.1), can focal length of camera not mediated, otherwise utilizes following formula to determine the adjusting time:

$t=\left\{\begin{array}{cc}\frac{\ln e}{\ln 6.119}+0.5217& e>1.1\\ \frac{\mathrm{<figure-callout\; id="1"\; label="ln"\; filenames="A20071001799200252.png"\; state="\{\{state\}\}">ln</figure-callout>}1/e}{\ln 6.119}+0.5217& e<0.9\end{array}\right.---\left(34\right)$

By this algorithm, just can realize regulating the purpose of focal length of camera according to Size Error, the needs that its regulating effect can satisfy people's observation get final product.Improve the control effect if desired, only need to adjust the error allowed band and get final product.

Claims (5)

1. trace and monitor method based on the video frequency motion target feature of dual camera linkage structure, by an overall view monitoring camera target of monitor area is discerned, determine the position of target, the speed of service and direction, after the locking tracking target, target information is passed to the feature that is turned to by cradle head control follow the tracks of camera, follow the tracks of camera by feature lock onto target is carried out feature, amplifying the back follows the tracks of, the feature picture of display-object, thereby obtain the target more information, it is characterized in that, this method is carried out according to the following steps

A. the feature that adopts an overall view monitoring camera and to be arranged on the The Cloud Terrace is followed the tracks of camera, adopt the interlock between a computer realization overall view monitoring camera and the feature tracking camera, promptly set up each some position corresponding relation in the captured video pictures of two cameras respectively in the scene picture, according to the position of target in the panoramic shooting head determine the feature camera towards, enable the target of following the tracks of over against needs;

B. utilizing the overall view monitoring camera that the target of monitor area is carried out behavior detects, promptly use the method for image and Video processing that moving target is separated with the background difference of scene, obtain the information of number, the speed of travel, direction of travel and the position of intrusion target;

C. to the above-mentioned detected Target Setting direction of motion, target location, movement velocity, four characteristic parameters of color of object, and note, the method of utilization template matches is followed the tracks of moving target between the frame of video and frame, obtains the motion track information of each target;

D. overall view monitoring camera interaction relation that the target trajectory information that obtains is set up by step a is transferred to feature and follows the tracks of camera, rotation by the control The Cloud Terrace drives the rotation of feature tracking camera, carry out target following, the target with locking is presented in the monitored picture after amplifying real-time and accurately;

E. control feature and follow the tracks of the convergent-divergent of the camera focal length target that furthers, the above-mentioned target in the monitored picture of being locked in that obtains is carried out feature and taken.

2. in accordance with the method for claim 1, it is characterized in that, among the described step b moving target being detected, is to adopt, and at first utilizes the pixel grey scale subsumption algorithm that background is carried out modeling, obtain a background frames, constantly this background is upgraded, adopted the frame-to-frame differences method again, moving object detection is gone out, concrete steps are as follows

At first, import one section video that includes the N two field picture, the pixel value of establishing current point be f (i, j), the pixel value x after the normalization (i j) is: x (i, j)=f (i, j)/255

Grey value profile to each pixel in this N two field picture is added up, and setting fluctuation range is 0.1, then the gray value after the normalization is divided into 10 value classes, to the array N of given one 10 dimension of each pixel value _{X (i, j)}(k), k=1,2 ..., 10, then have:

N _{X (i, j)}(k)=in the N two field picture, and gray value drop on (k-0.05, k+0.05) number of pixels in the scope }

Statistics is found out N after finishing _{X (i, j)}(k), k=1,2 ..., 10 when being maximum, corresponding k,

That is: N _Max=max{N _{X (i, j)}(k), k=1,2 ..., 10}

Calculate its average and variance, obtain the confidential interval of candidate background dot intensity profile, promptly obtained the background model of this monitoring image;

Afterwards, the point that is judged as the candidate target is carried out the specificity analysis of connected domain again, will be judged as the point deletion of non-target, what stay is exactly the point that is judged as target;

To the video pictures of a new frame of input, judge whether corresponding gray values of pixel points has dropped in the confidential interval, if do not drop in the confidential interval, this point is the pixel of target, at this moment, background model is constant; If drop in the confidential interval, then show this point for not by the background dot of target occlusion, then according to following formula computation of mean values and variance, revise background model;

Average is upgraded:

x _t+1(i，j)＝( x _t(i，j)+x _t+1(i，j))/2

Wherein, x _T+1(i, j) the pixel average of expression current time, x _t(i j) is the pixel average of previous moment, x _T+1(i j) is the pixel value of current time,

Variance is upgraded:

${{\mathrm{\&sigma;}}^2}_{t+1}(i,j)=\underset{k=1}{\overset{N_{\max }}{\mathrm{\&Sigma;}}}{(x(i,j)-{\stackrel{\&OverBar;}{x}}_{t+1}(i,j))}^2$

At this moment, be [x as the background frames image that detects moving target _T+1(i, j)] _M*n,

When carrying out monitor task, to the pixel value x of the present frame of input (i, j), adopt the method for frame-to-frame differences to calculate according to following formula:

e(i，j)＝|x(i，j)- x _t+1(i，j)|

If frame-to-frame differences e (i j) greater than prior preset threshold, shows that then this point is an impact point, otherwise, be indicated as background dot, with its with average more new formula or variance more new formula upgrade.

3. in accordance with the method for claim 1, it is characterized in that among the described step c, the method for utilization template matches is followed the tracks of concrete employing to moving target,

When not intersecting between the moving target, distinguish target by the distance between the detected target on the judgement two continuous frames, and target is numbered, draw the movement velocity and the direction of each target simultaneously;

When intersecting appears in the movement locus between a plurality of targets, then distinguish target, thereby obtain the movement locus of each target and it is carried out record by target speed, color characteristic, direction character are compared;

To the statistics of velocity to moving target, employing be that the method for the pixel count that moves in different frame of estimating target central point realizes;

The analysis of color of object feature then is by people's build and people's daily habits are divided into three parts to color characteristic promptly: head, and upper body, the lower part of the body is through an a certain proportion of average cutting apart and count three Color Channels of each part;

To the estimation of target travel direction, asking difference to obtain the direction of motion of target when the variation of former frame internal coordinates by target;

Distinguish the target of detection according to these three features then and it is numbered, obtain the motion track information of each moving target in real time.

4. in accordance with the method for claim 1, it is characterized in that, in the described steps d, adopt the MPC07 motion control card to send the rotation of commands for controlling The Cloud Terrace, be prestored into the movement function storehouse in the MPC07 motion control card and come the controlled motion mode, after detecting the intended target position, the pulse signal of positional information and control card is through certain transformational relation, control MPC07 sends the pulse signal of appointment, thereby drive feature tracking cam movement is finished the tracking to intended target.

5. in accordance with the method for claim 1, it is characterized in that among the described step e, following algorithm is adopted in the control of focusing:

The size of object is m in the video pictures, establishes given target expectation size and is M, and then error e is expressed as:

$e=\frac{m}{M}$

If error e is not mediated to focal length of camera, otherwise is utilized following formula to determine the adjusting time in the error allowed band of setting:

$t=\left\{\begin{array}{cc}\frac{\ln e}{\ln 6.119}+0.5217& e>1.1\\ \frac{\ln 1/e}{\ln 6.119}+0.5217& e<0.9\end{array}\right.$

The needs that its regulating effect can satisfy people's observation get final product.

Images