CN103516956B - Pan/Tilt/Zoom camera monitoring intrusion detection method - Google Patents

Abstract

A kind of Pan/Tilt/Zoom camera monitoring intrusion detection method, step one, obtain background frames, and set up the index of background frames: first video camera is resetted, then control video camera is from the beginning of initial angle, often gathers a two field picture and calculates the kinematic parameter between an adjacent image frame；Then, translational component according to the x, y, z direction obtained from kinematic parameter estimates the variable quantity that video camera Pan and Tilt rotates, when the accumulated change amount that video camera Pan and Tilt rotates exceedes setting numerical value, capture a width background frames, and with current this background frames of kinematic parameter labelling；Step 2, resets again by video camera, simultaneously using the background frames of gained as current background；Start after camera operation to rotate, often gather a two field picture and carry out a Pan and Tilt spin data estimation between present image and current background, background frames and present frame are registrated by Pan and the Tilt spin data according to obtaining, and utilize image background to remove method detection and swarm into object.

Description

Pan/Tilt/Zoom camera monitoring intrusion detection method

Technical field

The present invention relates to a kind of camera status detection method, invasion can be detected from camera acquisition image Object.

Background technology

During camera supervised, generally utilize background subtraction to determine whether what object was invaded.Select Which frame as current background frame it is to be appreciated that the current state of video camera.Camera status comprises two with one The vector representation of individual component, is to horizontally rotate angle, vertical tilt angle (Pan, Tilt) respectively.

Although the current state of video camera can be obtained from cradle head controllor, but to obtain accurate state value and need Cloud platform rotation to be stopped, because when The Cloud Terrace moves due to control structure machine error and communication delay, it is thus achieved that State value be not current last look.

Foundation and the maintenance of background model are particularly significant, more closely to intrusion detection method based on background subtraction Year researched and proposed a lot of background model and maintenance algorithm, but these algorithms to be both for video camera fixing Situation, the Background maintenance problem in the present invention is more complicated, in addition to considering the problem of general Background maintenance, It is also contemplated that the problem that image registration causes.

Summary of the invention

It is an object of the invention to a kind of simple camera status computational methods, can be used for background frames and select.

For achieving the above object, the present invention is by the following technical solutions:

A kind of Pan/Tilt/Zoom camera monitoring intrusion detection method, it comprises the steps:

Step one, obtains background frames, and sets up the index of background frames: is first resetted by video camera, then controls Video camera, from the beginning of initial angle, often gathers a two field picture and calculates the kinematic parameter between an adjacent image frame, Described kinematic parameter refers to represent Pan and the Tilt value of video camera attitude；Then, according to from kinematic parameter The translational component in the x, y, z direction obtained estimates the variable quantity that video camera Pan and Tilt rotates, and works as video camera When the accumulated change amount that Pan and Tilt rotates exceedes setting numerical value, capture a width background frames, and by current fortune Dynamic this background frames of parameter tags；

Step 2, detects the object swarming into video camera: again resetted by video camera, simultaneously by gained Background frames as current background；Start after camera operation rotate, often gather a two field picture carry out once when Pan and Tilt spin data between front image and current background is estimated, rotates according to Pan and Tilt obtained Background frames and present frame are registrated by data, utilize image background to remove method detection and swarm into object；If no Detect and swarm into object, then change background frames.

In described step 2, the anglec of rotation after camera operation is: adjacent two background image overlay regions Territory number is no less than 2/3rds of total image size.

In described step 2, use multiresolution layered image that background frames and present frame are registrated； And the generation of multiresolution layered image uses gaussian pyramid decomposition method.

In described step 2, described image background removes method and refers to: set up single Gaussian Background model, And combine the false detection that neighborhood relevance detection elimination is produced by image registration and background motion.

Use the present invention of technique scheme, it is possible to detect rapidly and accurately under camera motion state Invasion object, and this algorithm need not camera calibration, substantially increases detection efficiency.

Accompanying drawing explanation

Fig. 1 is the flow chart of the present invention.

Fig. 2 is background acquisition algorithm flow chart of the present invention.

Fig. 3 is background track algorithm flow chart of the present invention.

Fig. 4 is the one group of background frames captured on video camera pan path.

Fig. 5 is Moving Objects detection example the most in the same time.

Fig. 6 is the detection using background subtraction and elimination to obtain after registrating residual error and suppression falseness change pixel Result.Wherein Fig. 6 (1) background frames；Fig. 6 (2) present frame；Fig. 6 (3) present frame presses the figure after background frames registration Picture；Fig. 6 (4) directly background subtract after image；Fig. 6 (5) is that (4) are used threshold binarization treatment；Fig. 6 (6) The change pixel obtained after utilizing the false change algorithm of suppression.

Detailed description of the invention

Intrusion detection in the case of Pan/Tilt/Zoom camera rotary motion is in two stages: the first stage is pretreatment rank Section, completes capture background frames and set up index；Second stage is formal detection-phase, according to current camera Background frames is chosen in position, utilizes background to remove method and finds change pixel, as it is shown in figure 1, specifically include as follows Step:

Step one, obtains background frames, and sets up the index of background frames, this step mainly use image registration or Tracking technique calculates the camera motion of each background frames, and camera motion refers to represent video camera Pan and the Tilt value of attitude, but the present invention does not consider that Zoom and Roll moves, and is called for short PTZ value.First will Video camera resets, and then control video camera is from the beginning of initial angle, often gathers a two field picture and calculates the most adjacent Kinematic parameter between picture frame, then, according to the translation in the x, y, z direction obtained from kinematic parameter Component estimates the variable quantity that video camera Pan and Tilt rotates, when the accumulated change that video camera Pan and Tilt rotates When amount exceedes setting numerical value, capture a width background frames, and with current this background frames of kinematic parameter labelling.

Representing a three-dimensional coordinate point in initial camera coordinate system with (X, Y, Z), it is at two dimensional image The imaging point of coordinate system be (u, v), meets according to pin-hole imaging principle relation between them:

$\frac{u}{X}=\frac{v}{Y}=\frac{f}{Z}---\left(1\right)$

It is expressed as homogeneous coordinates form

$\left(\begin{array}{c}\mathrm{wx}\\ \mathrm{wy}\\ w\end{array}\right)=\left(\begin{array}{cccc}f& 0& 0& 0\\ 0& f& 0& 0\\ 0& 0& 1& 0\end{array}\right)\left(\begin{array}{c}X\\ Y\\ Z\\ 1\end{array}\right)---\left(2\right)$

Wherein f is lens focus.

If video camera is taken the photograph newly rotating around coordinate axes x, y, z rotation alpha, β, γ angle, point (X, Y, Z) The new coordinate (X ', Y ', Z ') of camera coordinate system, relation therebetween is

$\left[\begin{array}{c}{X}^{\′}\\ Y^{\′}\\ Z^{\′}\end{array}\right]=\left[\begin{array}{ccc}\cos \mathrm{\γ}& \sin \mathrm{\γ}& 0\\ -\sin \mathrm{\γ}& \cos \mathrm{\γ}& 0\\ 0& 0& 1\end{array}\right]\left[\begin{array}{ccc}\cos \mathrm{\β}& 0& -\sin \mathrm{\β}\\ 0& 1& 0\\ \sin \mathrm{\β}& 0& \cos \mathrm{\β}\end{array}\right]\left[\begin{array}{ccc}1& 0& 0\\ 0& \cos \mathrm{\α}& \sin \mathrm{\α}\\ 0& -\sin \mathrm{\α}& \cos \mathrm{\α}\end{array}\right]\left[\begin{array}{c}X\\ Y\\ Z\end{array}\right]---\left(3\right)$

When camera motion is less, and i.e. α, β, γ are the least, by trigonometric function approximate formula (sinx ≈ X, cosx ≈ 1) above formula is reduced to

$\left[\begin{array}{c}{X}^{\′}\\ Y^{\′}\\ Z^{\′}\end{array}\right]=\left[\begin{array}{ccc}1& \mathrm{\γ}& -\mathrm{\β}\\ -\mathrm{\γ}& 1& \mathrm{\α}\\ \mathrm{\β}& -\mathrm{\α}& 1\end{array}\right]\left[\begin{array}{c}X\\ Y\\ Z\end{array}\right]---\left(4\right)$

If new images coordinate is (x ', y '), then it is with (x, relation y) can be obtained by (2) (4)

$\left(\begin{array}{c}{w}^{\′}{x}^{\′}\\ w^{\′}{y}^{\′}\\ w^{\′}\end{array}\right)=\left(\begin{array}{ccc}{f}^{\′}& 0& 0\\ 0& f^{\′}& 0\\ 0& 0& 1\end{array}\right)\left(\begin{array}{c}{X}^{\′}\\ Y^{\′}\\ Z^{\′}\end{array}\right)=\left[\begin{array}{ccc}1& \mathrm{\γ}& -\mathrm{\β}\\ -\mathrm{\γ}& 1& \mathrm{\α}\\ \mathrm{\β}/f^{\′}& -\mathrm{\α}/f^{\′}& 1/f^{\′}\end{array}\right]\left[\begin{array}{c}{f}^{\′}\mathrm{wx}/f\\ f^{\′}\mathrm{wy}/f\\ f^{\′}w\end{array}\right]$

If using 4 parameter models, above formula can be write as

$\left\{\begin{array}{c}{x}^{\′}=m_{4}x+m_{3}x+m_1\\ y^{\′}=-m_{3}y+m_{4}y+m_2\end{array}\right.---\left(5\right)$

Wherein $m_1=-\frac{f^{\′}w}{{\mathrm{fw}}^{\′}}\mathrm{\β},m_2=\frac{f^{\′}w}{{\mathrm{fw}}^{\′}}\mathrm{\α},m_3=\frac{f^{\′}w}{{\mathrm{fw}}^{\′}}\mathrm{\γ},m4=\frac{f^{\′}w}{{\mathrm{fw}}^{\′}}$

Obviously the change of α, β, γ is reflected in m respectively₁、m₂、m₃In, if Jiao of video camera during monitoring Away from keeping constant, then So can be according to m₁、m₂、m₃Value approximate evaluation go out the value of α, β, γ, I.e. m₁、m₂、m₃Accumulated value (Pan, Tilt, Roll) can as background frames approximation index.Because not examining Consider the Roll motion of video camera, so Roll value is approximately 0.

Step 2, detects the object swarming into video camera: again resetted by video camera, simultaneously by step The background frames of one gained is as current background；Start after camera operation to rotate, often gather a two field picture and carry out One time Pan and the Tilt spin data between present image and current background is estimated, then according to the Pan obtained With Tilt spin data, background frames and present frame are registrated, utilize image background removing method detection to swarm into right As；If being not detected by swarming into object, then update the motion parameter of video camera, and then determine the need for Change background frames.When background frames need not change, kinematic parameter between present frame and background frames next time Estimation can utilize the result of last estimation quickly to calculate, because of referred to herein as image trace.When new background frames When changing, the kinematic parameter calculated between background frames and present frame for the first time is referred to as image registration, passes through Registration obtains kinematic parameter, the initiation parameter followed the tracks of as image motion parameters subsequently, until new Background frames is changed.

Specifically, after video camera resets, starting video camera, video camera often rotates to an angle generation one Background, the determination of the anglec of rotation is no less than the three of total image size with adjacent two background picture overlapping region numbers / bis-are advisable.

When Tilt takes a fixed value, the acquisition algorithm of background frames is as follows:

Step1: send reset command to cradle head controllor, waits that The Cloud Terrace has resetted

Step2: kinematic parameter initializes: t=0, ρ (t)=(m₁ m₂ m₃ m₄)=(0 00 0)；Pan initial value S=0, Pan old value S_old=0；Capture two field picture I (t), preserve kinematic parameter ρ (t) and present image I (t) as the first width Background frames；

Step3: capture next frame image I (t+1)；

Step4: calculate kinematic parameter change

δ ρ=-(H^tH)^-1 H^t(I(t+1)-I(t))；

S=S+ δ m₁；

Step5: update kinematic parameter ρ (t+1)=ρ (t)+δ ρ；

Step6: if W_sizeFor picture frame width, preserve kinematic parameter ρ (t+1) and currently Image I (t+1) is as background frames；S_old=S

Step7: if sign is (δ m₁)=sign (S), sign are symbol detection function, turn Step3, otherwise terminate.

Background track algorithm, for calculating the current rotation of video camera and angle of inclination (pan, tilt), selects with this Select suitable background picture and do background subtraction.When above-mentioned Tilt is fixed value, invasion object detection algorithm describes As follows:

Step1: send reset command to cradle head controllor, waits that The Cloud Terrace has resetted, captures sub-picture I (t)；

Step2: initialize camera motion ρ (t=0)=(m₁ m₂ m₃ m₄)=(0 00 0)；Read first width background Frame I_b；Initial pan value S=0；

Step3: capture next frame image I (t+1)；

Step4: calculate camera motion change

δ ρ=-(H^tH)^-1H^t(I(t+1)-I(t))；

Step5: update kinematic parameter S=S+ δ m₁；

Step6: retrieve most suitable background frames according to S, it may be judged whether need to change background, if needing to change Then read new background and give I_b, otherwise turn Step8；

Step7: use Multi-Resolution Registration technology to estimate kinematic parameter initial point ρ₀

ρ₀=esitmate (I (t+1), I_b)

Go to Step10；

Step8: estimate kinematic parameter

δρ₀=-(H^tH)^-1 H^t(I(t+1)-warp(ρ₀, I_b))；

Step9: update kinematic parameter ρ₀(t+1)=ρ₀(t)+δρ₀；

Step10: use background subtraction detection Moving Objects

Turn Step3.

Due to the background frames Limited Number prestored, the motion between present frame and background frames may be relatively big, now Direct solution equation, not only needs the calculating time grown very much, and likely has to the local optimum of equation Solving, general employing multiresolution layered image registration solves this problem.The life of multiresolution layered image One-tenth can use wavelet decomposition, it is possible to use gaussian pyramid decomposes, it is considered to the speed of algorithm and realization are multiple Polygamy, the present invention uses gaussian pyramid to decompose.Registration operation is from the beginning of lowest resolution, and low resolution obtains The kinematic parameter arrived, as the initial parameter of next high-resolution level, solves new more accurate kinematic parameter, Until highest resolution layer.

Actually at same stage resolution ratio, a registration operation is often difficult to obtain optimal solution, it is generally required to many Secondary registration circulation operation just can obtain optimal solution.Adjacent twice registration is similar to consecutive image motion problems, can Use the kinematic parameter model identical with layering registration.Repeatedly registration operation to same layers of resolution herein, Present frame is transformed under background frames coordinate system by the kinematic parameter utilizing previous step, in upper once circulation again Registrate with background frames, update kinematic parameter, so on, until no longer there is significant change in kinematic parameter Change.

In step 2, image background removes method and refers to: sets up single Gaussian Background model, and combines field Correlation detection eliminates the false detection produced by image registration and background motion.

Neighborhood relevance detection thought is, is that (x y), examines further for the some pixel of motion pixel to preliminary judgement Look into (whether x, y) meet surrounding pixels background model, if (x, y) belongs to surrounding pixels background model, then it is assumed that Owing to registration error or background motion cause, now it is no longer regarded as that (x y) is Moving Objects pixel, i.e.

Wherein M preserves the mark being judged to motion pixel point, Represent pixel (x, the collection of surrounding pixels y) Closing, (x y) is (x, y) intensity of position pixel, I in present frame to I_b(x y) is (x, y) position pixel in background frames Intensity, T (x, y) for change decision threshold.General detection neighborhood is set as the circle of a diameter of 3 to 5 pixels Region.

The renewal of background model: if (x y) is judged to that motion pixel then background model keeps constant, the most more New background model and decision threshold, i.e.

$I_b(x,y)=\left\{\begin{array}{c}\mathrm{\λ}{I}_b(x,y)+(1-\mathrm{\λ})I(x,y)\mathrm{ifM}(x,y)=1\\ I_b(x,y)\mathrm{otherwise}\end{array}\right.$

$T(x,y)=\left\{\begin{array}{c}\mathrm{\λT}(x,y)+4(1-\mathrm{\λ})|I(x,y)-I_b(x,y)|\mathrm{ifM}(x,y)=0\\ T(x,y)\mathrm{otherwise}\end{array}\right.$

Wherein λ ∈ [0,1] is renewal speed.

Intrusion detection experiment under camera motion, Fig. 4 is the one group of back of the body captured on video camera pan path Scape frame, Fig. 5 is one group of intrusion detection example, and Fig. 6 is for using background subtraction and eliminating registration residual error and press down The testing result obtained after system falseness change pixel.

Claims (4)

1. the camera supervised intrusion detection method of PTZ, it is characterised in that it comprises the steps:

Step one, obtain background frames, and set up the index of background frames: first video camera is resetted, then control video camera from the beginning of initial angle, often gathering a two field picture and calculate the kinematic parameter between an adjacent image frame, described kinematic parameter refers to represent Pan and the Tilt value of video camera attitude；Then, video camera Pan and Tilt is estimated according to the translational component in x, y direction obtained from kinematic parameter The variable quantity rotated, when the accumulated change amount that video camera Pan and Tilt rotates exceedes setting numerical value, captures a width background frames, and with current this background frames of kinematic parameter labelling；

Step 2, detects the object swarming into video camera: again resetted by video camera, simultaneously using the background frames of gained as current background；Start after camera operation to rotate, often gather a two field picture and carry out a Pan and Tilt spin data estimation between present image and current background, background frames and present frame are registrated by Pan and the Tilt spin data according to obtaining, and utilize image background to remove method detection and swarm into object；If being not detected by swarming into object, then change background frames.

2. according to the camera supervised intrusion detection method of the PTZ described in claim 1, it is characterised in that: in described step 2, determine the anglec of rotation after camera operation with adjacent two background image overlapping region numbers no less than 2/3rds of total image size.

3. according to the camera supervised intrusion detection method of the PTZ described in claim 1, it is characterised in that: in described step 2, use multiresolution layered image that background frames and present frame are registrated；And the generation of multiresolution layered image uses gaussian pyramid decomposition method.

4. according to the camera supervised intrusion detection method of the PTZ described in claim 1, it is characterized in that: in described step 2, described image background removes method and refers to: sets up single Gaussian Background model, and combines the false detection that neighborhood relevance detection elimination is produced by image registration and background motion.