CN101727570A - Tracking method, track detection processing unit and monitor system - Google Patents

Abstract

An embodiment of the invention discloses a tracking method, a track detection processing unit and a monitor system. The tracking method comprises: predicting the position of the movement of a target to obtain the predicted position of the target; extracting the prospect according to the obtained predicted position and determining a connected domain, making statistic to obtain the characteristic parameters of the connected domain; judging that the target characteristic update requirements are met by the result of the comparison between the characteristic parameters of the connected domain and the original characteristic parameters of the target, updating the original characteristic parameters of the target through the characteristic parameters of the connected domain matched during the comparison, and determining the position of the matched connected domain as the matched position for tracking the target. The invention is capable of well solving the problem that the characteristics of the target change in the process of tracking the target.

Description

Treatment facility and supervisory system are followed the tracks of in tracking, detection

Technical field

The present invention relates to the computer vision research technical field, be specifically related to a kind of tracking, detection tracking treatment facility and supervisory system.

Background technology

The objective self-adapting tracking is a research topic in the computer vision, how effectively interesting target effectively to be followed the tracks of, and be the gordian technique in the video monitoring system.Occurred many algorithm of target tracking at present, can roughly be divided into several classes: based on the track algorithm of " filtering, data allocations ", based on the track algorithm of " Target Modeling, location " with based on the track algorithm of " motion detection " etc.

(1) track algorithm based on " filtering, data allocations " comprises Kalman filtering algorithm, expanded Kalman filtration algorithm, particle filter algorithm etc.This type of track algorithm adopts the method for " state space " that discrete dynamic system is carried out modeling mostly.For example, Kalman filtering algorithm can be predicted next dbjective state constantly according to the historic state of target.(2) based on the track algorithm of " Target Modeling, location " comprise algorithm based on invariant moments, based on the algorithm of template matches and Mean-Shift algorithm (average drifting algorithm) etc.This class track algorithm generally is made up of three parts: the modeling of target, measuring similarity, search matched.Wherein, the Mean-Shift algorithm does not need to carry out exhaustive search, and the algorithm real-time is good, is a single characteristic parameter algorithm; Its adopts kernel function histogram-modeling, to the edge stop, the rotation of target, distortion be all insensitive, also is a kind of high performance pattern matching algorithm.(3) track algorithm based on motion detection comprises Pfinder algorithm and W ⁴Algorithms etc., this class track algorithm depends on motion detection algorithm.

In the process that target is followed the tracks of, can there be the target signature variation issue.

In the prior art,, can adopt W for the target signature variation issue ⁴Algorithm is followed the tracks of.W ⁴In the algorithm be adopt minimum, maximum intensity value and maximum time difference value be that each pixel is carried out statistical modeling in the scene, extract moving target, realization is to detection, the tracking of target, and the line period ground context update of going forward side by side therefore can processing target changing features problem on a certain degree.

In research and practice process to prior art, the inventor finds that there is following problem in prior art:

The W that prior art is adopted for the target signature variation issue ⁴Algorithm, be to rely on motion detection algorithm fully, when in prospect is extracted, more noise spot occurring, follow the tracks of frame violent variation can take place, if the situation of division takes place in the detected target of detection algorithm, then may follow the tracks of a target as two targets this moment, thereby and block mutually when causing the mutual adhesion of the prospect that extracts when moving target, can be used as a target to two targets this moment.

Therefore, prior art side does not have well to solve object appearing changing features problem in the process that target is followed the tracks of.

Summary of the invention

The embodiment of the invention provides a kind of tracking, detection to follow the tracks of treatment facility and supervisory system, can solve preferably target is carried out target signature variation issue in the tracing process.

According to an aspect of the present invention, provide a kind of tracking, comprising:

The position of target of prediction motion obtains the predicted position of described target;

Carry out foreground extraction according to the described predicted position that obtains, and definite connected domain, statistics obtains the characteristic parameter of connected domain;

The result who compares according to the characteristic parameter and the original characteristic parameter of described target of described connected domain judges and meets target signature more after the new demand, the characteristic parameter of the described connected domain that is matched during by comparison upgrades the original characteristic parameter of described target, and the position of determining the described connected domain that matches is as the matched position that target is followed the tracks of.

According to a further aspect in the invention, also provide a kind of detection to follow the tracks of treatment facility, comprising:

The target prodiction module is used for the position that target of prediction moves, and obtains the predicted position of described target;

Target detection and characteristics analysis module are used for carrying out foreground extraction according to the described predicted position that obtains, and definite connected domain, and statistics obtains the characteristic parameter of connected domain;

Target signature is upgraded decision-making module, is used for the result that characteristic parameter and the original characteristic parameter of described target according to described connected domain compare and judges whether to meet more new demand of target signature;

The target signature update module, being used for upgrading decision-making module in described target signature judges and meets target signature more after the new demand, the characteristic parameter of the described connected domain that is matched during by comparison upgrades the original characteristic parameter of described target, and the position of determining the described connected domain that matches is as the matched position that target is followed the tracks of.

Another aspect of the present invention also provides a kind of supervisory system, comprises the image acquisition input equipment and detects the tracking treatment facility, and described image acquisition input equipment is used for following the tracks of the image that the treatment facility input is gathered target to described detection; Described detection is followed the tracks of treatment facility and is comprised:

The target prodiction module is used for the image according to the target of described image acquisition input equipment input, and the position of target of prediction motion obtains the predicted position of described target;

Target detection and characteristics analysis module are used for carrying out foreground extraction according to the described predicted position that obtains, and definite connected domain, and statistics obtains the characteristic parameter of each connected domain;

Target signature is upgraded decision-making module, is used for the result that characteristic parameter and the original characteristic parameter of described target according to described connected domain compare and judges whether to meet more new demand of target signature;

The target signature update module, being used for upgrading decision-making module in described target signature judges and meets target signature more after the new demand, the characteristic parameter of the described connected domain that is matched during by comparison upgrades the original characteristic parameter of described target, and the position of determining the described connected domain that matches is as the matched position that target is followed the tracks of.

The technical scheme that the embodiment of the invention provides, owing to obtained the predicted position of target, be to utilize described predicted position when foreground extraction, and counted the characteristic parameter of each connected domain, the result who further compares according to the characteristic parameter and the original characteristic parameter of described target of described connected domain judges whether to meet more new demand of target signature then, meet target signature more after the new demand judging so, just can utilize the characteristic parameter of described connected domain that the original characteristic parameter of described target is upgraded, the clarification of objective parameter information is upgraded in time, thereby can obtain the latest features situation of target, so just can more effective maintenance to the tracking of target, avoid when detecting target division takes place, can following the tracks of a target as two targets or when moving target blocks mutually, can being used as two targets the problem appearance of a target.

Description of drawings

In order to be illustrated more clearly in the embodiment of the invention or technical scheme of the prior art, to do to introduce simply to the accompanying drawing of required use in embodiment or the description of the Prior Art below, apparently, accompanying drawing in describing below only is some embodiments of the present invention, for those of ordinary skills, under the prerequisite of not paying creative work, can also obtain other accompanying drawing according to these accompanying drawings.

Fig. 1 is the embodiment of the invention one a tracking process flow diagram;

Fig. 2 is the embodiment of the invention two tracking process flow diagrams;

Fig. 3 is the treatment scheme synoptic diagram that treatment facility is followed the tracks of in the single detection of the embodiment of the invention;

Fig. 4 is that the embodiment of the invention detects tracking treatment facility structural representation.

Embodiment

Below in conjunction with the accompanying drawing in the embodiment of the invention, the technical scheme in the embodiment of the invention is clearly and completely described, obviously, described embodiment only is the present invention's part embodiment, rather than whole embodiment.Based on the embodiment among the present invention, those of ordinary skills belong to the scope of protection of the invention not making the every other embodiment that is obtained under the creative work prerequisite.

The embodiment of the invention provides a kind of tracking, can solve preferably target is carried out target signature variation issue in the tracing process.The embodiment of the invention mainly by with several track algorithms in conjunction with utilization, and corresponding proposition decision mechanism and disposal route, thus more effectively solve the problem of the object appearing changing features in the process of monitoring and tracking target.

Seeing also Fig. 1, is the embodiment of the invention one tracking process flow diagram, comprises step:

The position of step 101, target of prediction motion obtains the predicted position of described target;

In this step, can adopt track algorithm, for example, adopt the position of Kalman filtering algorithm target of prediction motion wherein, obtain the predicted position of described target based on " filtering, data allocations ".

Step 102, carry out foreground extraction according to the described predicted position that obtains, and definite connected domain, statistics obtains the characteristic parameter of connected domain;

In this step, can adopt track algorithm based on motion detection.Carry out local foreground extraction according to the predicted position that step 101 obtains, obtain a bianry image, in this bianry image, carry out connected component labeling, determine connected domain, statistics obtains the characteristic parameter of each connected domain.Here said characteristic parameter comprises eigenwert probability distribution, pixel count, length breadth ratio, the dispersion degree of color of object, describes target by these characteristic parameters.

Step 103, the result who compares according to the characteristic parameter and the original characteristic parameter of described target of described connected domain judge and meet target signature more after the new demand, the characteristic parameter of the described connected domain that is matched during by comparison upgrades the original characteristic parameter of described target, and the position of determining the described connected domain that matches is as the matched position that target is followed the tracks of.

In this step, the result who compares according to the characteristic parameter and the original characteristic parameter of described target of described connected domain judge meet target signature more new demand comprise:

The determined coefficient of similarity value of eigenwert probability distribution according to the color of object of the eigenwert probability distribution of the color of object of each connected domain and described target matches a similar area;

Whether the pixel count, length breadth ratio, dispersion degree of judging described similar area belongs to setting range with the ratio of the pixel count of described target, length breadth ratio, dispersion degree respectively, if not, determine not meet more new demand of target signature, if determine to meet more new demand of target signature.

From this embodiment as can be seen, owing to obtained the predicted position of target, be to utilize described predicted position when foreground extraction, and counted the characteristic parameter of each connected domain, the result who further compares according to the characteristic parameter and the original characteristic parameter of described target of described connected domain judges whether to meet more new demand of target signature then, meet target signature more after the new demand judging so, just can utilize the characteristic parameter of described connected domain that the original characteristic parameter of described target is upgraded, the clarification of objective parameter information is upgraded in time, thereby can obtain the latest features situation of target, so just can more effective maintenance to the tracking of target, avoid in the prior art when detecting target division takes place, can following the tracks of a target as two targets or when moving target blocks mutually, can being used as two targets the problem appearance of a target.

Seeing also Fig. 2, is the embodiment of the invention two tracking process flow diagrams.Embodiment two more describes the tracking processing procedure in detail than embodiment one, comprises step among Fig. 2:

Step 201, adopt based on the track algorithm target of prediction of " filtering, data allocations " position at present frame;

Based on the track algorithm of " filtering, data allocations ", be that example describes but is not limited to this in this step to adopt Kalman filtering algorithm.

For video monitoring scene, the position of the target in the scene in each two field picture constituted the track of target travel.Adopt the purpose of Kalman filter in this step, be for possible position according to target in the prediction of the positional information before the target present frame, so state variable that obtains in the Kalman filter and observed reading are the positional information of target, the relevant information of the centre coordinate of more precisely tracked target.

For Kalman filtering algorithm, generally handle as follows:

Signal model: X (k)=A (k-1) X (k-1)+B (k) W (k) (1)

Observation model: Y (k)=C (k) X (k)+V (k) (2)

Wherein: X (k), Y (k) are respectively state vector and observation vector, A (k-1), B (k), C (k) are respectively state-transition matrix, input matrix, observing matrix, W (k) and V (k) are signal noise and observation noise, W (k) and V (k) are white Gaussian noise, the obedience average is 0 multivariate normal distribution, each component variance equates, is respectively σ _v=σ _w=5.

Suppose the motion of moving target center on X, Y-axis all be one by quickening at random by the rectilinear motion of disturbance, acceleration alpha is to satisfy a random quantity of Gaussian distribution, average is 0, variance is σ _w ², α satisfy α～N (0, σ _w ²) distribute acceleration alpha signal noise w (k) just here.

So, make signal vector X (k)=[x (k) y (k) x ' (k) y ' (k)] ^T, x (k) wherein, y (k) is respectively the location components of target's center on x, y axle, x ' is (k), y ' is respectively speed on x, y axle (k).Observation vector Y (k)=[x _c(k) y _c(k)] ^T, x wherein _c(k), y _c(k) be the observed reading of target's center position on X, Y-axis respectively, observation noise v (k) satisfy v (k)～N (0, σ _v ²) distribute.

According to above-mentioned definition, above-mentioned two models mentioning can be expressed as:

$\left[\begin{array}{c}x\left(k\right)\\ y\left(k\right)\\ x^{\′}\left(k\right)\\ y^{\′}\left(k\right)\end{array}\right]=\left[\begin{array}{cccc}1& 0& t& 0\\ 0& 1& 0& t\\ 0& 0& 1& 0\\ 0& 0& 0& 0\end{array}\right]\left[\begin{array}{c}x(k-1)\\ y(k-1)\\ x^{\′}(k-1)\\ y^{\′}(k-1)\end{array}\right]+\left[\begin{array}{c}{t}^2/2\\ t^2/2\\ t\\ t\end{array}\right]w\left(k\right)---\left(3\right)$

$\left[\begin{array}{c}{x}_c\left(k\right)\\ y_c\left(k\right)\end{array}\right]=\left[\begin{array}{cccc}1& 0& 0& 0\\ 0& 1& 0& 0\end{array}\right]\left[\begin{array}{c}x\left(k\right)\\ y\left(k\right)\\ x^{\′}\left(k\right)\\ y^{\′}\left(k\right)\end{array}\right]+\left[\begin{array}{c}1\\ 1\end{array}\right]v\left(k\right)---\left(4\right)$

Can set the constant empirical value is: t=1, σ _v=σ _w=5; And initial value X (1)=[x _Sy _S0 0] ^T, x wherein _S, y _SThe centre coordinate of target in the expression start frame.

With the value substitution of correlation parameter above-mentioned (3) and (4), then this step can obtain the position of target at present frame according to the position prediction before the target, promptly obtains a predicted position.

Step 202, employing are carried out the clarification of objective parametric statistics based on the track algorithm of motion detection;

Adopt track algorithm to carry out the clarification of objective parametric statistics in this step, before introducing this step in detail, introduce relevant characteristic parameter earlier and describe based on motion detection.

For the target of following the tracks of, can use the eigenwert probability distribution of color of object

Characteristic parameters such as pixel count C, length breadth ratio R, dispersion degree D are described target.Wherein, dispersion degree is defined as the area of object in the image and the ratio of girth, that is:

Be used to represent the degree of scatter of target object on image.

Below introduce the eigenwert probability distribution of color of object in detail

Hypothetical target is centered close to

Wherein there be n pixel to use

Expression, the pixel color rgb value is m through the number of the eigenwert bin after quantizing, and bin is meant an interval quantity of value, and then the eigenwert probability distribution of this color of object can be expressed as:

${\hat{q}}_u=\mathrm{Const}*\underset{i=1}{\overset{n}{\mathrm{\Σ}}}k\left({\left|\right|\frac{{\stackrel{\→}{x}}_i-{\stackrel{\→}{x}}_0}{h}\left|\right|}^2\right)\mathrm{\δ}[b\left({\stackrel{\→}{x}}_i\right)-u]---\left(5\right)$

Wherein, k (x) is the profile function of gaussian kernel function, k (x)=e ^xOwing to block or the influence of background, near the pixel the object module center is more reliable than peripheral pixel, and k (x) is to distribute big weights to the pixel at center, and is to distribute little weights for deep pixel.Among the function k (x)

Effect be influence in order to eliminate different big or small targets and to calculate, the target of ellipse representation is normalized to a unit circle.δ (x) is the Delta function,

Be pixel color value, Total effect be to judge position in the target area The pixel color quantized value whether belong between u chromatic zones, if belong to, then Value is 1, otherwise is 0.Const is a standardized constant factor, makes Therefore:

$\mathrm{Const}=\frac{1}{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}k\left({\left|\right|\frac{{\stackrel{\→}{x}}_i-{\stackrel{\→}{x}}_0}{h}\left|\right|}^2\right)}---\left(6\right)$

After the description introduction of relevant feature parameters is intact, below introduce the specific operation process of this step:

1) target is carried out local foreground extraction;

The algorithm that carries out foreground extraction for target in this step can be to adopt based on any one target detection in the track algorithm of motion detection and foreground extraction algorithm, for example background subtraction algorithm, mixed Gauss model algorithm, figure partitioning algorithm or the like.

Because moving target has the continuity in time and space between adjacent two frames, therefore need in the scope of the overall situation, not carry out motion target detection in this step, be benchmark only with the predicted position that obtains through step 201, in the subrange of 1.5 to 2 times of target sizes of this place-centric, carry out just passablely, so just accelerated the speed of moving object detection greatly.

Use the mixed Gauss model algorithm in the embodiment of the invention,, target is carried out local foreground extraction, obtain a bianry image according to existing mixed Gauss model algorithm process process.

2) the provincial characteristics parametric statistics of target signature.

Above-mentionedly obtain a bianry image, in this bianry image, carried out connected component labeling so.The connected domain here is to adopt four mode of communicating, and for two pixels, if there is a place to link to each other in the four direction of upper and lower, left and right, then two points are connectivity points, and connected domain is the set of the point that is interconnected.

Through behind the connected component labeling, obtain n connected domain { O _iI=1 ... n, the central point of each connected domain is

At this moment, carry out the provincial characteristics parametric statistics of target signature, comprise the pixel count C that adds up each connected domain _i, length breadth ratio R _i, dispersion degree D _iWith the eigenwert probability distribution of statistics original image at the color of object of each connected domain

According to the formula in above-mentioned (5), can obtain

${\hat{p}}_{\mathrm{ui}}\left({\stackrel{\→}{y}}_i\right)=\mathrm{Const}*\underset{j=1}{\overset{n}{\mathrm{\Σ}}}k\left({\left|\right|\frac{{\stackrel{\→}{x}}_j-{\stackrel{\→}{y}}_i}{h}\left|\right|}^2\right)\mathrm{\δ}[b\left({\stackrel{\→}{x}}_j\right)-u]---\left(7\right)$

Through 1) and 2) processing, then step 202 has obtained the ASSOCIATE STATISTICS value of target signature parameter.

Step 203, judge whether to meet more new demand of target signature, if, enter step 204, if not, enter step 206;

In this step, suppose that the color feature value probability distribution of tracking target is

Pixel count is C _Target, length breadth ratio R _Target, dispersion degree D _Target

At first, utilize tracking target With the eigenwert probability distribution of original image at the color of object of each connected domain Match the highest regional O of similarity _i

With

Similarity with Bhattacharrya coefficient (coefficient of similarity is also referred to as Pasteur's coefficient)

Measure distribution, promptly

$\widehat{\mathrm{\ρ}}\left(\stackrel{\→}{y}\right)\≡\mathrm{\ρ}[p\left(\stackrel{\→}{y}\right),q]=\underset{u=1}{\overset{m}{\mathrm{\Σ}}}\sqrt{{\hat{p}}_{\mathrm{ui}}\left(\stackrel{\→}{y}\right){\hat{q}}_{\mathrm{ut}\arg \mathrm{et}}}---\left(8\right)$

Coefficient of similarity

Big more, show that then similarity is high more, thereby, draw the highest regional O of similarity according to each comparative result _i

Then, according to the highest O of similarity that matches _iThe zone, check O _iThe pixel count C in zone _i, length breadth ratio R _i, dispersion degree D _iCompare whether within an acceptable scope Deng the character pair parameter of characteristic parameter and target.If satisfy following three conditions, then be true, otherwise be false.

${\mathrm{\&sigma;}}_C<\frac{C_i}{C_{t\arg \mathrm{et}}}<1/{\mathrm{\&sigma;}}_C$

${\mathrm{\&sigma;}}_R<\frac{R_i}{R_{t\arg \mathrm{et}}}<1/{\mathrm{\&sigma;}}_R$

${\mathrm{\&sigma;}}_D<\frac{D_i}{D_{t\arg \mathrm{et}}}<1/{\mathrm{\&sigma;}}_D$

Wherein, σ _CBe O _iProportion threshold value is counted, σ with object pixel in the zone _DBe O _iZone and target dispersion degree proportion threshold value, σ _RBe O _iZone and target length breadth ratio proportion threshold value.σ _C, σ _D, σ _RCan be the value between the 0-1, specifically the different values of setting can be arranged according to different situations.σ in case study on implementation of the present invention _C, σ _D, σ _RIn all value be 0.6, can certainly get other values.

If O _iThe character symbol in zone closes and states decision mechanism, then assert O _iBe the position of tracking target, determine to meet more new demand of target signature, enter next step and adopt O _iThe feature in zone is upgraded the characteristic parameter of tracking target.

Step 204, carry out target signature and upgrade; According to the decision mechanism in the step 203, assert O _iBe tracking target, determine to meet target signature more after the new demand, in this step, adopt O _iThe feature in zone is upgraded the characteristic parameter of tracking target, also is equivalent to simultaneously carry out the self-adaptation adjustment to following the tracks of frame, and concrete operations are as follows:

${\hat{q}}_{\mathrm{ut}\arg \mathrm{et}}={\mathrm{\&gamma;}}_Q*{\hat{p}}_{\mathrm{ui}}\left({\stackrel{\&RightArrow;}{y}}_i\right)+(1-{\mathrm{\&gamma;}}_Q)*{\hat{q}}_{\mathrm{ut}\arg \mathrm{et}}$

C _target＝γ _C*C _i+(1-γ _C)*C _target

R _target＝γ _R*R _i+(1-γ _R)*R _target

D _target＝γ _D*D _i+(1-γ _D)*D _target

Wherein, γ _Q, γ _C, γ _R, γ _DBe respectively and follow the tracks of the color feature value probability distribution

Pixel count C _Target, length breadth ratio R _Target, dispersion degree D _TargetTurnover rate.In case study on implementation of the present invention with γ _QValue is 0.05, can be with γ _C, γ _RAnd γ _DAll value is 0.2, can certainly get other values.

Through above-mentioned processing, the clarification of objective parameter information is upgraded in time, just can obtain the latest features situation of target, thereby can more effective maintenance to the tracking of target, avoid in the prior art when detecting target division takes place, can following the tracks of a target as two targets or when moving target blocks mutually, can being used as two targets the problem appearance of a target.

Step 205, determine that the connected domain position that matches is the matched position of target, enter step 209;

Because determined to meet more new demand of target signature, and be to adopt O _iThe feature in zone is upgraded the characteristic parameter of tracking target, therefore determines that the connected domain that matches is O _iThe zone is the matched position of target, and this position is the optimal location of tracking, enters step 209 then and carries out tracking results output.

Need to prove there is not inevitable ordinal relation between the step 204 and 205.

Step 206, be starting point, adopt and follow the tracks of, enter step 207 based on the track algorithm of " Target Modeling, location " with the predicted position;

According to the decision mechanism in the step 203, if do not assert O _iBeing tracking target, promptly not meeting target signature more after the new demand, then is starting point with the predicted position, adopts and follows the tracks of based on the track algorithm of " Target Modeling, location ".

Adopt track algorithm in this step, illustrate with the Mean-Shift algorithm but be not limited to this based on " Target Modeling, location ".

For making the Bhattacharrya coefficient

Maximum promptly will be sought the optimal objective position in present frame, the predicted position that obtains with employing Kalman filtering prediction in the step 201 in this step is as the center of the position of present frame search window

Seek the optimal objective position in the neighborhood

Mean-Shift algorithm iteration process is as follows:

If target has The position of Kalman filtering prediction is positioned at

The size of target area is h, repeats following steps so and can obtain the target reposition

A, estimate in the present frame with formula (5)

The eigenwert probability distribution of the color of place's candidate target

B, use formula

Each point in the zoning

Weights { ω _iI=1 ... n.

C, use Mean-Shift algorithm, calculate the target reposition:

${\stackrel{\&RightArrow;}{y}}_1=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{\stackrel{\&RightArrow;}{x}}_i{\mathrm{\&omega;}}_{i}g\left[{\left|\right|\frac{{\stackrel{\&RightArrow;}{y}}_0-{\stackrel{\&RightArrow;}{x}}_i}{h}\left|\right|}^2\right]}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{\mathrm{\&omega;}}_{i}g\left[{\left|\right|\frac{{\stackrel{\&RightArrow;}{y}}_0-{\stackrel{\&RightArrow;}{x}}_i}{h}\left|\right|}^2\right]}$

In the formula, g (x) is for being similarly the profile function of kernel function, and g (x)=-k ' (x)=-e ^x

If D Then stop to calculate, otherwise

Change steps A, wherein choosing of ε should make

With

Between the distance less than a pixel.

After through the Mean-Shift algorithm convergence, obtain target area O _m, target area O _mCentral point be

Step 207, judge whether similarity meets similar requirement, if, enter step 208, if not, enter step 209;

The target area O that in step 206, obtains _mAfter, calculate O according to formula (5) _mThe color feature value probability distribution be

Calculate according to formula (8)

With

The Bhattacharrya coefficient

This coefficient tolerance With

Similarity.

Judge this moment

With

Similarity whether meet similar requirement, specifically pass through

Compare with pre-set threshold and to obtain, this threshold value can be provided with different values as the case may be:

If

Greater than threshold value, expression meets similar requirement, enters step 208, if

Less than threshold value, expression does not meet similar requirement, enters step 209.

Step 208, to follow the tracks of the tracing positional that obtains based on the track algorithm of " Target Modeling, location " be the matched position of target to adopt, and enters step 210;

Because

Greater than threshold value, be to meet similar requirement, therefore determine in this step that adopting the Mean-Shift algorithm to follow the tracks of has obtained correct result, that is to say that following the tracks of the position that obtains with employing based on the track algorithm of " Target Modeling, location " is the matched position of target, this position is the optimal location of tracking, enters step 210 then and carries out tracking results output.

Step 209, be the matched position of target, enter step 210 with the predicted position;

Because

Less than threshold value, be not meet similar requirement, therefore the definite employing of this step Mean-Shift algorithm is followed the tracks of and is not obtained correct result, and possible reason is that target is blocked or blocked fully by large tracts of land, and the result who at this moment adopts the Mean-Shift algorithm to follow the tracks of is insecure.And under general situation, the motion of target is continuously, clocklike, do not have very strong maneuverability, so the predicted position that draws through Kalman filtering is relatively reliable.Therefore, exist

Under the situation less than threshold value, the predicted position that definite directly employing Kalman filtering draws is as the target location, and this position is the optimal location of tracking, enters step 210 then and carries out tracking results output.

Step 210, according to the matched position output tracking result of target.

In this step, the matched position output tracking result of the target that obtains according to above-mentioned steps, thus effectively keep tracking to target.

From this embodiment as can be seen, this embodiment has been integrated use several track algorithms, utilized the advantage separately of these track algorithms, at first be utilize to adopt based on the track algorithm target of prediction of " filtering; data allocations " position at present frame, so just obtained the predicted position of target, then be to adopt track algorithm to advance the clarification of objective parametric statistics based on motion detection, and a kind of decision mechanism proposed, promptly behind the characteristic parameter that has counted each connected domain, the result who compares according to the characteristic parameter and the original characteristic parameter of described target of described connected domain judges whether to meet more new demand of target signature, meet target signature more after the new demand if judge, just can utilize the characteristic parameter of described connected domain that the original characteristic parameter of described target is upgraded, such renewal just can keep target is followed the tracks of more accurately and effectively, do not meet the target signature update request if judge, can further utilize this moment based on the track algorithm of " Target Modeling; location " and follow the tracks of and obtain tracing area, and according to similarity determine relatively that to adopt tracing area still be the predicted position that obtains before adopting as the matched position of target, also keep effective tracking more accurately thereby distinguish different situations to target.

The technical scheme of the embodiment of the invention can be used for the supervisory system to fixed scene, such as parking lot monitoring, bank monitoring, building monitoring or the like.Usually in the supervisory system, comprise image acquisition input equipment (as video camera), detect and follow the tracks of treatment facility (computing machine or embedded device); If be the collaborative supervisory system of multiple-camera, control server in also comprising, the network equipments such as switch.The technical scheme of the embodiment of the invention specifically is that the detection that can be used for supervisory system is followed the tracks of in the treatment facility, can realize that following the tracks of frame changes along with the size variation of target, and clarification of objective upgraded in time, realizes the target tracking of longer time.

In concrete the application, no matter be the supervisory system or the collaborative supervisory system of multiple-camera of single camera, follow the tracks of treatment facility for single detection, processing procedure all is the same, below follows the tracks of treatment facility with single detection and uses the treatment scheme of embodiment of the invention technical scheme and illustrate.

Seeing also Fig. 3, is the treatment scheme synoptic diagram that treatment facility is followed the tracks of in the single detection of the embodiment of the invention.

As shown in Figure 3, detect the tracking treatment facility and comprise image capture module M01, target selection and analysis module M02, target prodiction module M03, target detection and characteristics analysis module M04, target signature renewal decision-making module M05, target signature update module M06, tracking and target judging module M07, output module M08.The following description of cooperation processing procedure between each module:

Image capture module M01, being used for obtaining the output result that target is carried out obtaining after the image acquisition from image acquisition input equipment (for example video capture device such as video camera or video frequency collection card) is video sequence S01.S01 is if the first frame video needs through target selection and analysis module M02, otherwise directly sends to target prodiction module M03.

Target selection and analysis module M02 can also can be finished by computing machine according to the initial setting condition by the monitoring staff according to the manually selected tracking target of actual conditions automatically to the selection of target.After target selection is finished, target selection and analysis module M02 also need target signature is analyzed, obtain characteristic parameters such as target location, color of object eigenwert probability distribution, pixel count, length breadth ratio, dispersion degree, the S02 as a result of output comprises these characteristic parameters.

Target prodiction module M03 can adopt Kalman filtering algorithm, according to the possible position of target in the information prediction present frame of target location point in the past.At video sequence is under the situation of first frame, and target prodiction module M03 is according to the position of the positional information target of prediction of target among the S02 as a result of output; And be not under the situation of first frame at video sequence, according to the position at the position feature target of prediction place in the tracking results (S07 or S08) of former frame.Except the predicted position of target, also comprise S02 as a result, the S07 of output or the characteristic parameter information of the target signature among the S08 among the S03 as a result of target prodiction module M03 output.

Target detection and characteristics analysis module M04, carry out local foreground extraction near the future position in the S03 as a result of output in the field and carry out connected component labeling, and calculate the center position of each connected domain, the color feature value probability distribution of adding up pixel count, length breadth ratio, dispersion degree and the connected domain zone of each connected domain.Comprise the above-mentioned characteristic parameter of each connected domain among the S04 as a result of target detection and characteristics analysis module M04 output, and comprised the characteristic parameter information among the S03 as a result that exports.

Target signature is upgraded decision-making module M05, at first according to the color feature value probability distribution of the tracking target among the S04 as a result of output and original image in the eigenwert probability distribution of the color of object of each connected domain, match the highest regional O of similarity _i, check O then _iWhether the character pair parameter of characteristic parameter such as the pixel count, length breadth ratio, dispersion degree in zone and target is compared within a receivable scope.If satisfy the condition of setting, then be true, show O _iThe feature in zone meets the decision mechanism that target signature is upgraded decision-making module M05, assert O _iBe tracking target, export S05 as a result to target signature update module M06; Otherwise be false, show O _iThe feature in zone does not meet the decision mechanism that target signature is upgraded decision-making module M05, does not assert O _iBe tracking target, follow the tracks of and target judging module M07 output S06 to Mean-Shift.The characteristic parameter information that has all still comprised S04 among the S05 as a result of output and the S06.

Target signature update module M06 is with O among the S05 as a result of output _iThe characteristic parameter in zone upgrades the feature of tracking target, the characteristic parameter that upgrades comprises color of object eigenwert probability distribution, pixel count, length breadth ratio and dispersion degree, export S07 as a result to output module M08 then, wherein comprised the position of target, each characteristic parameter after the renewal.

Follow the tracks of and target judging module M07, can adopt the Mean-Shift algorithm to carry out relevant treatment and obtain the target location.Tracking and target judging module M07 are the center with the position of Kalman filtering prediction, adopt the Mean-Shift algorithm to carry out iteration and follow the tracks of.After the convergence of Mean-Shift track algorithm, obtain target area Om, calculate the similarity of Om and target.If similarity Bhattacharrya coefficient

Greater than threshold value, expression Mean-Shift track algorithm has obtained correct result.If less than threshold value, expression Mean-Shift track algorithm has not obtained correct result, and at this moment the Mean-Shift tracking results is that the predicted position that insecure definite Kalman filtering draws is reliable relatively.Therefore, exist

Under the situation less than threshold value, directly adopt predicted position that Kalman filtering draws,, wherein comprised target location that tracking and target judging module M07 obtain and the clarification of objective parameter among the S06 to output module M08 output S08 as a result as the target location.

Output module M08 is according to the result of target signature update module M06, tracking and target judging module M07 input, outwards output tracking result.

In this Application Example, carrying out above-mentioned flow process, can effectively solve target is carried out target signature variation issue in the tracing process by detect following the tracks of treatment facility.

The introduction that foregoing is detailed the tracking of the embodiment of the invention, corresponding, below the detailed detection of introducing the embodiment of the invention and providing follow the tracks of treatment facility and supervisory system.

Seeing also Fig. 4, is that the embodiment of the invention detects tracking treatment facility structural representation.

As shown in Figure 4, detecting the tracking treatment facility comprises: target prodiction module 11, target detection and characteristics analysis module 12, target signature are upgraded decision-making module 13, target signature update module 14.

Target prodiction module 11 is used for the position that target of prediction moves, and obtains the predicted position of described target.

Target detection and characteristics analysis module 12 are used for carrying out foreground extraction according to the described predicted position that obtains, and definite connected domain, and statistics obtains the characteristic parameter of connected domain.

Target signature is upgraded decision-making module 13, is used for the result that characteristic parameter and the original characteristic parameter of described target according to described connected domain compare and judges whether to meet more new demand of target signature.

Target signature update module 14, being used for upgrading decision-making module 13 in described target signature judges and meets target signature more after the new demand, the characteristic parameter of the described connected domain that is matched during by comparison upgrades the original characteristic parameter of described target, and the position of determining the described connected domain that matches is as the matched position that target is followed the tracks of.

Detecting the tracking treatment facility also comprises: follow the tracks of and target judging module 15.

Follow the tracks of and target judging module 15, being used for upgrading decision-making module 13 in described target signature judges and does not meet target signature more after the new demand, with described predicted position is starting point, adopt the track algorithm of based target modeling, location to obtain tracing area, statistics obtains the characteristic parameter of tracing area, and according to the characteristic parameter of described tracing area and result that the original characteristic parameter of described target compares judge meet similar requirement after, determine with described tracing area as the matched position that target is followed the tracks of.

Described tracking and target judging module 15 comprise: tracking and statistic unit 151, judging unit 152, first processing unit 153, second processing unit 154.

Follow the tracks of and statistic unit 151, being used for upgrading decision-making module 13 in described target signature judges and does not meet target signature more after the new demand, with described predicted position is starting point, adopts the track algorithm of based target modeling, location to obtain tracing area, and statistics obtains the characteristic parameter of tracing area.

Judging unit 152 is used for judging whether to meet similar requirement according to the characteristic parameter of described tracing area to the result that the original characteristic parameter of described target compares.

First processing unit 153, be used for described judging unit 152 judge meet similar requirement after, determine with described tracing area as the matched position that target is followed the tracks of.

Second processing unit 154, be used for described judging unit 152 judge do not meet similar requirement after, determine with described predicted position as the matched position that target is followed the tracks of.

The characteristic parameter of the top connected domain of mentioning comprises eigenwert probability distribution, pixel count, length breadth ratio, the dispersion degree of color of object.

Described target signature is upgraded decision-making module 13 and is comprised: matching unit 131, decision package 132.

Matching unit 131 is used for matching a similar area according to the determined coefficient of similarity value of eigenwert probability distribution of the color of object of the eigenwert probability distribution of the color of object of each connected domain and described target.

Decision package 132, whether the pixel count, length breadth ratio, dispersion degree that is used to judge described similar area belongs to setting range with the ratio of the pixel count of described target, length breadth ratio, dispersion degree respectively, if not, determine not meet more new demand of target signature, if determine to meet more new demand of target signature.

The characteristic parameter of the top tracing area of mentioning is the eigenwert probability distribution of color of object.

Judging unit 152 in described tracking and the target judging module 15 comprises: comparing unit 1521, determining unit 1522 as a result.

Comparing unit 1521, whether the determined coefficient of similarity value of eigenwert probability distribution of color of object that is used for the eigenwert probability distribution of color of object of comparison tracing area and described target is greater than setting threshold.

Determining unit 1522 as a result, be used for when described comparing unit 1521 compares described coefficient of similarity value less than setting threshold, determine not meet similar requirement, when described comparing unit 1521 compares described coefficient of similarity value greater than setting threshold, determine to meet similar requirement.

The embodiment of the invention also provides a kind of supervisory system, comprise the image acquisition input equipment and detect the tracking treatment facility, the image acquisition input equipment is used for following the tracks of the image that the treatment facility input is gathered target to described detection, described detection is followed the tracks of treatment facility and is followed the tracks of processing according to the image of the target of image acquisition input equipment input, the concrete structure of described detection tracking treatment facility no longer is described in detail as shown in Figure 4 herein.

In sum, the technical scheme that the embodiment of the invention provides, owing to obtained the predicted position of target, be to utilize described predicted position when foreground extraction, and counted the characteristic parameter of each connected domain, the result who further compares according to the characteristic parameter and the original characteristic parameter of described target of described connected domain judges whether to meet more new demand of target signature then, meet target signature more after the new demand judging so, just can utilize the characteristic parameter of described connected domain that the original characteristic parameter of described target is upgraded, the clarification of objective parameter information is upgraded in time, thereby can obtain the latest features situation of target, so just can more effective maintenance to the tracking of target, avoid when detecting target generation division, can following the tracks of a target as two targets in the prior art, when moving target blocks mutually, can be used as two targets the problem appearance of a target.

Further, embodiment of the invention technical scheme judge do not meet the target signature update request after, can further utilize to follow the tracks of and obtain tracing area based on the track algorithm of " Target Modeling, location ", and according to similarity determine relatively that to adopt tracing area still be the predicted position that obtains before adopting as the matched position of target, also keep effective tracking more accurately thereby distinguish different situations to target.

More than a kind of tracking that the embodiment of the invention provided, detection are followed the tracks of treatment facility and supervisory system is described in detail, used specific case herein principle of the present invention and embodiment are set forth, the explanation of above embodiment just is used for helping to understand method of the present invention and core concept thereof; Simultaneously, for one of ordinary skill in the art, according to thought of the present invention, the part that all can change in specific embodiments and applications, in sum, this description should not be construed as limitation of the present invention.

Claims (12)

1. a tracking is characterized in that, comprising:

The position of target of prediction motion obtains the predicted position of described target;

Carry out foreground extraction according to the described predicted position that obtains, and definite connected domain, statistics obtains the characteristic parameter of connected domain;

The result who compares according to the characteristic parameter and the original characteristic parameter of described target of described connected domain judges and meets target signature more after the new demand, the characteristic parameter of the described connected domain that is matched during by comparison upgrades the original characteristic parameter of described target, and the position of determining the described connected domain that matches is as the matched position that target is followed the tracks of.

2. tracking according to claim 1 is characterized in that, described method also comprises:

The result who compares according to the characteristic parameter and the original characteristic parameter of described target of described connected domain judges and does not meet target signature more after the new demand, with described predicted position is starting point, adopt the track algorithm of based target modeling, location to obtain tracing area, statistics obtains the characteristic parameter of tracing area;

According to the characteristic parameter of described tracing area and result that the original characteristic parameter of described target compares judge meet similar requirement after, determine with described tracing area as the matched position that target is followed the tracks of.

3. tracking according to claim 2 is characterized in that, described method also comprises:

Described characteristic parameter according to described tracing area and the result that the original characteristic parameter of described target compares determine with described predicted position as the matched position that target is followed the tracks of after judging and not meeting similar requirement.

4. according to each described tracking of claim 1 to 3, it is characterized in that:

The characteristic parameter of described connected domain comprises eigenwert probability distribution, pixel count, length breadth ratio, the dispersion degree of color of object;

The described result who compares according to the characteristic parameter and the original characteristic parameter of described target of described connected domain judge meet target signature more the step of new demand comprise:

The determined coefficient of similarity value of eigenwert probability distribution according to the color of object of the eigenwert probability distribution of the color of object of each connected domain and described target matches a similar area;

Whether the pixel count, length breadth ratio, dispersion degree of judging described similar area belongs to setting range with the ratio of the pixel count of described target, length breadth ratio, dispersion degree respectively, if not, determine not meet more new demand of target signature, if determine to meet more new demand of target signature.

5. according to claim 2 or 3 described trackings, it is characterized in that:

The characteristic parameter of described tracing area is the eigenwert probability distribution of color of object;

Described characteristic parameter according to described tracing area is judged the step that meets similar requirement to the result that the original characteristic parameter of described target compares and is comprised:

Whether the determined coefficient of similarity value of eigenwert probability distribution of color of object of judging the eigenwert probability distribution of color of object of tracing area and described target if not, determines do not meet similar requirement, if determine meet similar requirement greater than setting threshold.

6. one kind is detected the tracking treatment facility, it is characterized in that, comprising:

The target prodiction module is used for the position that target of prediction moves, and obtains the predicted position of described target;

Target detection and characteristics analysis module are used for carrying out foreground extraction according to the described predicted position that obtains, and definite connected domain, and statistics obtains the characteristic parameter of connected domain;

Target signature is upgraded decision-making module, is used for the result that characteristic parameter and the original characteristic parameter of described target according to described connected domain compare and judges whether to meet more new demand of target signature;

The target signature update module, being used for upgrading decision-making module in described target signature judges and meets target signature more after the new demand, the characteristic parameter of the described connected domain that is matched during by comparison upgrades the original characteristic parameter of described target, and the position of determining the described connected domain that matches is as the matched position that target is followed the tracks of.

7. treatment facility is followed the tracks of in detection according to claim 6, it is characterized in that, also comprises:

Follow the tracks of and the target judging module, being used for upgrading decision-making module in described target signature judges and does not meet target signature more after the new demand, with described predicted position is starting point, adopt the track algorithm of based target modeling, location to obtain tracing area, statistics obtains the characteristic parameter of tracing area, and according to the characteristic parameter of described tracing area and result that the original characteristic parameter of described target compares judge meet similar requirement after, determine with described tracing area as the matched position that target is followed the tracks of.

8. treatment facility is followed the tracks of in detection according to claim 7, it is characterized in that described tracking and target judging module comprise:

Follow the tracks of and statistic unit, being used for upgrading decision-making module in described target signature judges and does not meet target signature more after the new demand, with described predicted position is starting point, adopts the track algorithm of based target modeling, location to obtain tracing area, and statistics obtains the characteristic parameter of tracing area;

Judging unit is used for judging whether to meet similar requirement according to the characteristic parameter of described tracing area to the result that the original characteristic parameter of described target compares;

First processing unit is used for after described judgment unit judges goes out to meet similar requirement, determines with described tracing area as the matched position that target is followed the tracks of;

Second processing unit is used for after described judgment unit judges goes out not meet similar requirement, determines with described predicted position as the matched position that target is followed the tracks of.

9. treatment facility is followed the tracks of in each described detection according to claim 6 to 8, it is characterized in that, the characteristic parameter of described connected domain comprises eigenwert probability distribution, pixel count, length breadth ratio, the dispersion degree of color of object, and described target signature is upgraded decision-making module and comprised:

Matching unit is used for matching a similar area according to the determined coefficient of similarity value of eigenwert probability distribution of the color of object of the eigenwert probability distribution of the color of object of each connected domain and described target;

Decision package, whether the pixel count, length breadth ratio, dispersion degree that is used to judge described similar area belongs to setting range with the ratio of the pixel count of described target, length breadth ratio, dispersion degree respectively, if not, determines not meet more new demand of target signature, if determine to meet more new demand of target signature.

10. treatment facility is followed the tracks of in detection according to claim 8, it is characterized in that the characteristic parameter of described tracing area is the eigenwert probability distribution of color of object, and the judging unit in described tracking and the target judging module comprises:

Whether comparing unit, the determined coefficient of similarity value of eigenwert probability distribution of color of object that is used for the eigenwert probability distribution of color of object of comparison tracing area and described target greater than setting threshold,

Determining unit is used for determining not meet similar requirement when described comparing unit compares described coefficient of similarity value less than setting threshold as a result, when described comparing unit compares described coefficient of similarity value greater than setting threshold, determines to meet similar requirement.

11. a supervisory system is characterized in that, comprises the image acquisition input equipment and detects the tracking treatment facility, described image acquisition input equipment is used for following the tracks of the image that the treatment facility input is gathered target to described detection, and described detection is followed the tracks of treatment facility and comprised:

The target prodiction module is used for the image according to the target of described image acquisition input equipment input, and the position of target of prediction motion obtains the predicted position of described target;

Target detection and characteristics analysis module are used for carrying out foreground extraction according to the described predicted position that obtains, and definite connected domain, and statistics obtains the characteristic parameter of each connected domain;

Target signature is upgraded decision-making module, is used for the result that characteristic parameter and the original characteristic parameter of described target according to described connected domain compare and judges whether to meet more new demand of target signature;

The target signature update module, being used for upgrading decision-making module in described target signature judges and meets target signature more after the new demand, the characteristic parameter of the described connected domain that is matched during by comparison upgrades the original characteristic parameter of described target, and the position of determining the described connected domain that matches is as the matched position that target is followed the tracks of.

12. supervisory system according to claim 11 is characterized in that, described detection is followed the tracks of treatment facility and is also comprised:

Follow the tracks of and the target judging module, being used for upgrading decision-making module in described target signature judges and does not meet target signature more after the new demand, with described predicted position is starting point, adopts the track algorithm of based target modeling, location to obtain tracing area, and statistics obtains the characteristic parameter of tracing area; According to the characteristic parameter of described tracing area and result that the original characteristic parameter of described target compares judge meet similar requirement after, determine with described tracing area as the matched position that target is followed the tracks of.

Images