CN106296730A - A kind of Human Movement Tracking System - Google Patents

Abstract

The invention discloses a kind of Human Movement Tracking System, including human motion video acquisition device, image preprocess apparatus, shooting adjusting apparatus and follow-up mechanism；Wherein, described video image is carried out processing obtaining by described follow-up mechanism follows the tracks of object present frame position, it was predicted that follow the tracks of object motion direction, according to described tracking object present frame position, and determine area-of-interest, at area-of-interest, described tracking object is tracked；Described shooting adjusting apparatus for judge described tracking object present frame position whether in the central area of current picture, if it is, do not adjust photographic head, if it does not, according to described tracking object motion direction adjust photographic head.The tracking effect that the present invention is capable of smoothing need not any auxiliary locator, is not limited by following the tracks of angle, can follow the tracks of human body in all directions, and robustness is not by external influence.

Description

A kind of Human Movement Tracking System

Technical field

The present invention relates to human body tracking technical field, be specifically related to a kind of Human Movement Tracking System.

Background technology

In correlation technique, mans motion simulation refers to detect human body position in the image gathered, and controls to take the photograph accordingly As the motion of head, tracking object is made to remain in the middle of picture.Correlation technique use infrared detection technique and ultrasound wave visit Survey technologies etc. carry out mans motion simulation, and wherein infrared detection technique uses and follows the tracks of subject wears infrared launcher, video camera The infrared signal received according to infrared receiving device, determines the camera site of video camera.But infrared track there is problems of resisting Interference is poor, is easily affected by thermal light sources such as the visible ray in environment, daylight lamp, and needs auxiliary device, turns when following the tracks of object The loss of infrared signal can be caused, it is impossible to judge shooting direction, the accuracy of impact location and real-time during body.Ultrasonic listening Multiple ultrasonic emitting with particular frequencies and ultrasonic probe, ultrasonic receiver are arranged near tracking object by technology, according to super The echo change that acoustic receiver device receives judges to follow the tracks of the position of object, determines the direction of video camera shooting.Due to ultrasonic The transmitting angle changing rate of ripple is big, and the angle therefore shooting orientation is the highest, it is impossible to follow the tracks of target for 360 degree.It addition, this technology also needs Want auxiliary equipment, and long Ultrasonic Radiation is to human health.

Summary of the invention

For solving the problems referred to above, it is desirable to provide a kind of Human Movement Tracking System.

The purpose of the present invention realizes by the following technical solutions:

A kind of Human Movement Tracking System, adjusts including human motion video acquisition device, image preprocess apparatus, shooting Device and follow-up mechanism, described human motion video acquisition device is for obtaining the video image comprising human body；Described image is pre- Processing means, for the video image gathered carries out pretreatment, eliminates the impact of video jitter；Described follow-up mechanism is to described Video image carries out processing obtaining follows the tracks of object present frame position, it was predicted that follow the tracks of object motion direction, according to described tracking object Present frame position, and determine area-of-interest, at area-of-interest, described tracking object is tracked；Described shooting adjusts dress Put for judge described tracking object present frame position whether in the central area of current picture, if it is, do not adjust shooting Head, if it does not, adjust photographic head according to described tracking object motion direction.

The invention have the benefit that, by selected tracking object, bonding position prediction adjusts the position of photographic head, it is achieved Smooth tracking effect, the present invention need not any auxiliary locator, do not limited by following the tracks of angle, can follow the tracks of in all directions Human body, robustness is not by external influence, thus solves above-mentioned technical problem.

Accompanying drawing explanation

The invention will be further described to utilize accompanying drawing, but the application scenarios in accompanying drawing does not constitute any limit to the present invention System, for those of ordinary skill in the art, on the premise of not paying creative work, it is also possible to obtain according to the following drawings Other accompanying drawing.

Fig. 1 is the structural representation of the present invention；

Fig. 2 is the module connection diagram of follow-up mechanism of the present invention.

Reference:

Human motion video acquisition device 1, image preprocess apparatus 2, shooting adjusting apparatus 3, follow-up mechanism 4, interested Area determination module 41, candidate motion region extraction module 42, tracking object-location 43, initialization submodule 421, state Metastasis model sets up submodule 422, observation model sets up submodule 423, candidate motion region calculating sub module 424, position are repaiied Syndrome generation module 425, resampling submodule 426.

Detailed description of the invention

In conjunction with following application scenarios, the invention will be further described.

Application scenarios 1

See Fig. 1, Fig. 2, the Human Movement Tracking System under the complex scene of an embodiment of this application scene, including Human motion video acquisition device 1, image preprocess apparatus 2, shooting adjusting apparatus 3 and follow-up mechanism 4, described human motion regards Frequently harvester 1 is for obtaining the video image comprising human body；Described image preprocess apparatus 2 is for the video image gathered Carry out pretreatment, eliminate the impact of video jitter；Described video image is carried out processing obtaining by described follow-up mechanism 4 follows the tracks of object Present frame position, it was predicted that follow the tracks of object motion direction, according to described tracking object present frame position, and determine area-of-interest, At area-of-interest, described tracking object is tracked；Described shooting adjusting apparatus 3 is used for judging that described tracking object is current Whether frame position is in the central area of current picture, if it is, do not adjust photographic head, if it does not, according to described tracking object The direction of motion adjusts photographic head.

Preferably, described carry out described video image processes acquisition tracking object present frame position, including: to described figure The candidate motion region comprising human body is extracted as carrying out processing；In described candidate motion region, obtain human body target；According to institute State human body target, determine tracking object, obtain and record tracking object present frame position；According to the described current framing bit of tracking object Put the described tracking object motion direction of prediction.

The above embodiment of the present invention adjusts the position of photographic head by selected object of following the tracks of, bonding position prediction, it is achieved flat Sliding tracking effect, the present invention need not any auxiliary locator, do not limited by following the tracks of angle, can follow the tracks of people in all directions Body, robustness is not by external influence, thus solves above-mentioned technical problem.

Preferably, the described video image to gathering carries out pretreatment, including: the first frame figure of selected described video image Picture is reference frame, and reference frame is averagely divided into four regions of non-overlapping copies, and W represents the width of image, and H represents figure image height Degree, four area size are 0.5W × 0.5H, start from image upper left according to being followed successively by region 1,2,3,4 clockwise； At the image center location selection area A that next frame receives₀, A₀Size be chosen to be 0.5W × 0.5H；By A₀According to above-mentioned side Method is divided into four image subblock A that size is 0.25W × 0.25H₁、A₂、A₃、A₄, A₁And A₂For estimating in vertical direction Local motion vector, A₃And A₄For estimating the local motion vector in horizontal direction, make A₁、A₂、A₃、A₄Respectively 1,2,3,4 Search optimal coupling in four regions, thus estimate the global motion vector of video sequence, then carry out inverse motion compensation, Eliminate the impact of video jitter.

Video image is carried out steady as processing by this preferred embodiment, it is to avoid successive image is processed and to cause by video jitter Impact, the efficiency of pretreatment is high.

Preferably, described follow-up mechanism 4 includes that area-of-interest determines module 41, candidate motion region extraction module 42 and Follow the tracks of object-location 43；Described area-of-interest determines that module 41 is emerging for determining sense in a two field picture of video image Interest region D₁And in this, as To Template；Described candidate motion region extraction module 42 is used for setting up particle state transfer and seeing Survey model and based on above-mentioned model, use particle filter predicting candidate moving region；Described tracking object-location 43 is used for Described candidate motion region and described To Template are carried out feature similarity tolerance, recognition and tracking object, and records tracking object Present frame position.

This preferred embodiment constructs the module architectures of follow-up mechanism 4.

Preferably, described candidate motion region extraction module 42 includes:

(1) initialization submodule 421: at described area-of-interest D₁Inside randomly select particle that quantity is n right Each particle carries out initialization process, and after initialization process, the original state of particle is x₀ ⁱ, initial weight is { Q_o ⁱ=1/n, i= 1,...n}；

(2) state transition model sets up submodule 422: be used for setting up particle state metastasis model, and described particle state turns Shifting formwork type employing following formula:

$x_m^i={\mathrm{Ax}}_{m-1}^i+v_m^i$

In formula,The new particle in expression m moment, m >=2,Being the white Gaussian noise of 0 for average, A is 4 rank unit matrix； The particle in m-1 moment is propagated by state transition model；

(3) observation model sets up submodule 423, for by color histogram, textural characteristics rectangular histogram and movement edge The mode that feature combines sets up particle observation model；

(4) candidate motion region calculating sub module 424: it utilizes minimum variance estimate to calculate candidate motion region:

$x_{now}=\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_m^j\·x_m^j$

In formula, x_nowRepresent the candidate motion region of the current frame image calculated,Represent the correspondence of m moment jth particle State value；

(5) position correction submodule 425: be used for revising abnormal data:

$x_{pre}=\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_{m-1}^j\·x_{m-1}^j$

In formula, x_preRepresent the candidate motion region of the current frame image calculated,Represent m-1 moment jth particle Corresponding states value；

Data exception evaluation function P=is set | x_now-x_pre|, if the value of P is more than the empirical value T, then x set_now=x_pre；

(6) resampling submodule 426: for deleting, by re-sampling operations, the particle that weights are too small, during resampling, utilize The difference of the prediction of system current time and observation provides and newly ceases residual error, and then carries out the particle of sampling by measuring new breath residual error Online adaptive adjusts, and in sampling process, the contextual definition between number of particles and information residual error is:

Wherein, N_mRepresent the number of particles in m moment, N in sampling process_maxAnd N_minRepresent minimum and maximum particle respectively Number, N_min+1Represent and be merely greater than N_minPopulation, N_max-1Represent and be only smaller than N_maxPopulation,The new breath of etching system when representing m Residual error.

This preferred embodiment uses the side combined based on color histogram, textural characteristics rectangular histogram and motion edge character Formula carries out the right value update of sampling particle, effectively enhances the robustness of tracking system；Position correction submodule 425, energy are set Enough avoid the impact that whole system is brought by abnormal data；In resampling submodule 426, utilize current time prediction and observation Difference provide and newly cease residual error, and then by measuring new breath residual error, the particle of sampling is carried out online adaptive adjustment, and fixed Relation between number of particles and information residual error in justice sampling process, preferably ensure that high efficiency and the algorithm of particle sampler Real-time.

Preferably, the particle right value update formula of described particle observation model is:

$Q_m^j=\frac{\stackrel{\‾}{Q_{Cm}^j}\·\stackrel{\‾}{Q_{Mm}^j}\·\stackrel{\‾}{Q_{Wm}^j}+{\mathrm{\λ}}_1\stackrel{\‾}{Q_{Cm}^j}+{{\mathrm{\λ}}_2}^2\stackrel{\‾}{Q_{Mm}^j}+{{\mathrm{\λ}}_2}^3\stackrel{\‾}{Q_{Wm}^j}+{\mathrm{\λ}}_1{\mathrm{\λ}}_2{\mathrm{\λ}}_3}{(1+{\mathrm{\λ}}_1)(1+{\mathrm{\λ}}_2)(1+{\mathrm{\λ}}_3)}$

In formula

$\stackrel{\‾}{Q_{Cm}^j}=Q_{Cm}^j/\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_{Cm}^j,Q_{Cm}^j=Q_{C(m-1)}^j\frac{1}{\sqrt{2\mathrm{\π}\mathrm{\σ}}}\exp (-\frac{A_m^2}{2{\mathrm{\σ}}^2})$

$\stackrel{\‾}{Q_{Mm}^j}=Q_{Mm}^j/\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_{Mm}^j,Q_{Mm}^j=Q_{M(m-1)}^j\frac{1}{\sqrt{2\mathrm{\π}\mathrm{\σ}}}\exp (-\frac{B_m^2}{2{\mathrm{\σ}}^2})$

$\stackrel{\‾}{Q_{Wm}^j}=Q_{Wm}^j/\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_{Wm}^j,Q_{Wm}^j=Q_{W(m-1)}^j\frac{1}{\sqrt{2\mathrm{\π}\mathrm{\σ}}}\exp (-\frac{C_m^2}{2{\mathrm{\σ}}^2})$

Wherein,Represent the final updated weights of m moment jth particle,WithRepresent m moment and m-1 respectively Jth particle renewal based on color histogram weights in moment,Represent jth in m moment and m-1 moment Particle renewal based on movement edge weights,Represent that in m moment and m-1 moment, jth particle is special based on texture Levy histogrammic renewal weights, A_mFor between the observation based on color histogram of jth particle in the m moment and actual value Bhattacharrya distance, B_mFor between the observation based on movement edge of jth particle in the m moment and actual value Bhattacharrya distance, C_mFor jth particle in the m moment based between the histogrammic observation of textural characteristics and actual value Bhattacharrya distance, σ is Gauss likelihood model variance, λ₁Normalized for feature weight based on color histogram The self-adaptative adjustment factor, λ₂For the normalized self-adaptative adjustment factor of feature weight based on movement edge, λ₃For special based on texture Levy the histogrammic feature weight normalized self-adaptative adjustment factor；

The computing formula of the described self-adaptative adjustment factor is:

${\mathrm{\λ}}_s^m={\mathrm{\ξ}}_{m-1}\·\[-\underset{j=1}{\overset{n}{\mathrm{\Σ}}}\left(p_{m-1}^{s/j}\right){\log }_2^{p_{m-1}^{s/j}}\],s=1,2,3;$

Wherein, during s=1,Represent in the m moment the normalized self-adaptative adjustment of feature weight based on color histogram because of Son,For the eigenvalue based on color histogram in m-1 moment observation probability value under j particle；During s=2,Represent The normalized self-adaptative adjustment factor of feature weight based on movement edge in the m moment,For in the m-1 moment based on motion limit The eigenvalue of edge observation probability value under j particle；During s=3,Represent in the m moment based on the histogrammic spy of textural characteristics Levy the weights normalized self-adaptative adjustment factor,For in the m-1 moment based on the histogrammic eigenvalue of textural characteristics at j grain Observation probability value under Zi；ξ_m-1Represent the locus variance yields of all particles in the m-1 moment.

This preferred embodiment proposes particle right value update formula and the calculating of the self-adaptative adjustment factor of particle observation model Formula, carries out fusion treatment to the feature weight of particle, effectively overcomes the defect that additivity merges and the property taken advantage of fusion exists, enters one Step enhances the robustness of tracking system.

In this application scenarios, choosing population n=50, tracking velocity improves 8% relatively, and tracking accuracy improves relatively 7%.

Application scenarios 2

See Fig. 1, Fig. 2, the Human Movement Tracking System under the complex scene of an embodiment of this application scene, including Human motion video acquisition device 1, image preprocess apparatus 2, shooting adjusting apparatus 3 and follow-up mechanism 4, described human motion regards Frequently harvester 1 is for obtaining the video image comprising human body；Described image preprocess apparatus 2 is for the video image gathered Carry out pretreatment, eliminate the impact of video jitter；Described video image is carried out processing obtaining by described follow-up mechanism 4 follows the tracks of object Present frame position, it was predicted that follow the tracks of object motion direction, according to described tracking object present frame position, and determine area-of-interest, At area-of-interest, described tracking object is tracked；Described shooting adjusting apparatus 3 is used for judging that described tracking object is current Whether frame position is in the central area of current picture, if it is, do not adjust photographic head, if it does not, according to described tracking object The direction of motion adjusts photographic head.

Preferably, described carry out described video image processes acquisition tracking object present frame position, including: to described figure The candidate motion region comprising human body is extracted as carrying out processing；In described candidate motion region, obtain human body target；According to institute State human body target, determine tracking object, obtain and record tracking object present frame position；According to the described current framing bit of tracking object Put the described tracking object motion direction of prediction.

The above embodiment of the present invention adjusts the position of photographic head by selected object of following the tracks of, bonding position prediction, it is achieved flat Sliding tracking effect, the present invention need not any auxiliary locator, do not limited by following the tracks of angle, can follow the tracks of people in all directions Body, robustness is not by external influence, thus solves above-mentioned technical problem.

Preferably, the described video image to gathering carries out pretreatment, including: the first frame figure of selected described video image Picture is reference frame, and reference frame is averagely divided into four regions of non-overlapping copies, and W represents the width of image, and H represents figure image height Degree, four area size are 0.5W × 0.5H, start from image upper left according to being followed successively by region 1,2,3,4 clockwise； At the image center location selection area A that next frame receives₀, A₀Size be chosen to be 0.5W × 0.5H；By A₀According to above-mentioned side Method is divided into four image subblock A that size is 0.25W × 0.25H₁、A₂、A₃、A₄, A₁And A₂For estimating in vertical direction Local motion vector, A₃And A₄For estimating the local motion vector in horizontal direction, make A₁、A₂、A₃、A₄Respectively 1,2,3,4 Search optimal coupling in four regions, thus estimate the global motion vector of video sequence, then carry out inverse motion compensation, Eliminate the impact of video jitter.

Video image is carried out steady as processing by this preferred embodiment, it is to avoid successive image is processed and to cause by video jitter Impact, the efficiency of pretreatment is high.

Preferably, described follow-up mechanism 4 includes that area-of-interest determines module 41, candidate motion region extraction module 42 and Follow the tracks of object-location 43；Described area-of-interest determines that module 41 is emerging for determining sense in a two field picture of video image Interest region D₁And in this, as To Template；Described candidate motion region extraction module 42 is used for setting up particle state transfer and seeing Survey model and based on above-mentioned model, use particle filter predicting candidate moving region；Described tracking object-location 43 is used for Described candidate motion region and described To Template are carried out feature similarity tolerance, recognition and tracking object, and records tracking object Present frame position.

This preferred embodiment constructs the module architectures of follow-up mechanism 4.

Preferably, described candidate motion region extraction module 42 includes:

(1) initialization submodule 421: at described area-of-interest D₁Inside randomly select particle that quantity is n right Each particle carries out initialization process, and after initialization process, the original state of particle is x₀ ⁱ, initial weight is { Q_o ⁱ=1/n, i= 1,...n}；

(2) state transition model sets up submodule 422: be used for setting up particle state metastasis model, and described particle state turns Shifting formwork type employing following formula:

$x_m^i={\mathrm{Ax}}_{m-1}^i+v_m^i$

In formula,The new particle in expression m moment, m >=2,Being the white Gaussian noise of 0 for average, A is 4 rank unit matrix；m- The particle in 1 moment is propagated by state transition model；

(3) observation model sets up submodule 423, for by color histogram, textural characteristics rectangular histogram and movement edge The mode that feature combines sets up particle observation model；

(4) candidate motion region calculating sub module 424: it utilizes minimum variance estimate to calculate candidate motion region:

$x_{now}=\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_m^j\·x_m^j$

In formula, x_nowRepresent the candidate motion region of the current frame image calculated,Represent the correspondence of m moment jth particle State value；

(5) position correction submodule 425: be used for revising abnormal data:

$x_{pre}=\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_{m-1}^j\·x_{m-1}^j$

In formula, x_preRepresent the candidate motion region of the current frame image calculated,Represent m-1 moment jth particle Corresponding states value；

Data exception evaluation function P=is set | x_now-x_pre|, if the value of P is more than the empirical value T, then x set_now=x_pre；

(6) resampling submodule 426: for deleting, by re-sampling operations, the particle that weights are too small, during resampling, utilize The difference of the prediction of system current time and observation provides and newly ceases residual error, and then carries out the particle of sampling by measuring new breath residual error Online adaptive adjusts, and in sampling process, the contextual definition between number of particles and information residual error is:

Wherein, N_mRepresent the number of particles in m moment, N in sampling process_maxAnd N_minRepresent minimum and maximum particle respectively Number, N_min+1Represent and be merely greater than N_minPopulation, N_max-1Represent and be only smaller than N_maxPopulation,The new breath of etching system when representing m Residual error.

This preferred embodiment uses the side combined based on color histogram, textural characteristics rectangular histogram and motion edge character Formula carries out the right value update of sampling particle, effectively enhances the robustness of tracking system；Position correction submodule 425, energy are set Enough avoid the impact that whole system is brought by abnormal data；In resampling submodule 426, utilize current time prediction and observation Difference provide and newly cease residual error, and then by measuring new breath residual error, the particle of sampling is carried out online adaptive adjustment, and fixed Relation between number of particles and information residual error in justice sampling process, preferably ensure that high efficiency and the algorithm of particle sampler Real-time.

Preferably, the particle right value update formula of described particle observation model is:

$Q_m^j=\frac{\stackrel{\‾}{Q_{Cm}^j}\·\stackrel{\‾}{Q_{Mm}^j}\·\stackrel{\‾}{Q_{Wm}^j}+{\mathrm{\λ}}_1\stackrel{\‾}{Q_{Cm}^j}+{{\mathrm{\λ}}_2}^2\stackrel{\‾}{Q_{Mm}^j}+{{\mathrm{\λ}}_2}^3\stackrel{\‾}{Q_{Wm}^j}+{\mathrm{\λ}}_1{\mathrm{\λ}}_2{\mathrm{\λ}}_3}{(1+{\mathrm{\λ}}_1)(1+{\mathrm{\λ}}_2)(1+{\mathrm{\λ}}_3)}$

In formula

$\stackrel{\‾}{Q_{Cm}^j}=Q_{Cm}^j/\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_{Cm}^j,Q_{Cm}^j=Q_{C(m-1)}^j\frac{1}{\sqrt{2\mathrm{\π}\mathrm{\σ}}}\exp (-\frac{A_m^2}{2{\mathrm{\σ}}^2})$

$\stackrel{\‾}{Q_{Mm}^j}=Q_{Mm}^j/\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_{Mm}^j,Q_{Mm}^j=Q_{M(m-1)}^j\frac{1}{\sqrt{2\mathrm{\π}\mathrm{\σ}}}\exp (-\frac{B_m^2}{2{\mathrm{\σ}}^2})$

$\stackrel{\‾}{Q_{Wm}^j}=Q_{Wm}^j/\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_{Wm}^j,Q_{Wm}^j=Q_{W(m-1)}^j\frac{1}{\sqrt{2\mathrm{\π}\mathrm{\σ}}}\exp (-\frac{C_m^2}{2{\mathrm{\σ}}^2})$

Wherein,Represent the final updated weights of m moment jth particle,WithRepresent m moment and m-1 respectively Jth particle renewal based on color histogram weights in moment,Represent jth in m moment and m-1 moment Particle renewal based on movement edge weights,Represent that in m moment and m-1 moment, jth particle is special based on texture Levy histogrammic renewal weights, A_mFor between the observation based on color histogram of jth particle in the m moment and actual value Bhattacharrya distance, B_mFor between the observation based on movement edge of jth particle in the m moment and actual value Bhattacharrya distance, C_mFor jth particle in the m moment based between the histogrammic observation of textural characteristics and actual value Bhattacharrya distance, σ is Gauss likelihood model variance, λ₁Normalized for feature weight based on color histogram The self-adaptative adjustment factor, λ₂For the normalized self-adaptative adjustment factor of feature weight based on movement edge, λ₃For special based on texture Levy the histogrammic feature weight normalized self-adaptative adjustment factor；

The computing formula of the described self-adaptative adjustment factor is:

${\mathrm{\λ}}_s^m={\mathrm{\ξ}}_{m-1}\·\[-\underset{j=1}{\overset{n}{\mathrm{\Σ}}}\left(p_{m-1}^{s/j}\right){\log }_2^{p_{m-1}^{s/j}}\],s=1,2,3;$

Wherein, during s=1,Represent in the m moment the normalized self-adaptative adjustment of feature weight based on color histogram because of Son,For the eigenvalue based on color histogram in m-1 moment observation probability value under j particle；During s=2,Represent The normalized self-adaptative adjustment factor of feature weight based on movement edge in the m moment,For in the m-1 moment based on motion limit The eigenvalue of edge observation probability value under j particle；During s=3,Represent in the m moment based on the histogrammic spy of textural characteristics Levy the weights normalized self-adaptative adjustment factor,For in the m-1 moment based on the histogrammic eigenvalue of textural characteristics at j grain Observation probability value under Zi；ξ_m-1Represent the locus variance yields of all particles in the m-1 moment.

This preferred embodiment proposes particle right value update formula and the calculating of the self-adaptative adjustment factor of particle observation model Formula, carries out fusion treatment to the feature weight of particle, effectively overcomes the defect that additivity merges and the property taken advantage of fusion exists, enters one Step enhances the robustness of tracking system.

In this application scenarios, choosing population n=55, tracking velocity improves 7% relatively, and tracking accuracy improves relatively 8%.

Application scenarios 3

See Fig. 1, Fig. 2, the Human Movement Tracking System under the complex scene of an embodiment of this application scene, including Human motion video acquisition device 1, image preprocess apparatus 2, shooting adjusting apparatus 3 and follow-up mechanism 4, described human motion regards Frequently harvester 1 is for obtaining the video image comprising human body；Described image preprocess apparatus 2 is for the video image gathered Carry out pretreatment, eliminate the impact of video jitter；Described video image is carried out processing obtaining by described follow-up mechanism 4 follows the tracks of object Present frame position, it was predicted that follow the tracks of object motion direction, according to described tracking object present frame position, and determine area-of-interest, At area-of-interest, described tracking object is tracked；Described shooting adjusting apparatus 3 is used for judging that described tracking object is current Whether frame position is in the central area of current picture, if it is, do not adjust photographic head, if it does not, according to described tracking object The direction of motion adjusts photographic head.

Preferably, described carry out described video image processes acquisition tracking object present frame position, including: to described figure The candidate motion region comprising human body is extracted as carrying out processing；In described candidate motion region, obtain human body target；According to institute State human body target, determine tracking object, obtain and record tracking object present frame position；According to the described current framing bit of tracking object Put the described tracking object motion direction of prediction.

The above embodiment of the present invention adjusts the position of photographic head by selected object of following the tracks of, bonding position prediction, it is achieved flat Sliding tracking effect, the present invention need not any auxiliary locator, do not limited by following the tracks of angle, can follow the tracks of people in all directions Body, robustness is not by external influence, thus solves above-mentioned technical problem.

Preferably, the described video image to gathering carries out pretreatment, including: the first frame figure of selected described video image Picture is reference frame, and reference frame is averagely divided into four regions of non-overlapping copies, and W represents the width of image, and H represents figure image height Degree, four area size are 0.5W × 0.5H, start from image upper left according to being followed successively by region 1,2,3,4 clockwise； At the image center location selection area A that next frame receives₀, A₀Size be chosen to be 0.5W × 0.5H；By A₀According to above-mentioned side Method is divided into four image subblock A that size is 0.25W × 0.25H₁、A₂、A₃、A₄, A₁And A₂For estimating in vertical direction Local motion vector, A₃And A₄For estimating the local motion vector in horizontal direction, make A₁、A₂、A₃、A₄Respectively 1,2,3,4 Search optimal coupling in four regions, thus estimate the global motion vector of video sequence, then carry out inverse motion compensation, Eliminate the impact of video jitter.

Video image is carried out steady as processing by this preferred embodiment, it is to avoid successive image is processed and to cause by video jitter Impact, the efficiency of pretreatment is high.

Preferably, described follow-up mechanism 4 includes that area-of-interest determines module 41, candidate motion region extraction module 42 and Follow the tracks of object-location 43；Described area-of-interest determines that module 41 is emerging for determining sense in a two field picture of video image Interest region D₁And in this, as To Template；Described candidate motion region extraction module 42 is used for setting up particle state transfer and seeing Survey model and based on above-mentioned model, use particle filter predicting candidate moving region；Described tracking object-location 43 is used for Described candidate motion region and described To Template are carried out feature similarity tolerance, recognition and tracking object, and records tracking object Present frame position.

This preferred embodiment constructs the module architectures of follow-up mechanism 4.

Preferably, described candidate motion region extraction module 42 includes:

(1) initialization submodule 421: at described area-of-interest D₁Inside randomly select particle that quantity is n right Each particle carries out initialization process, and after initialization process, the original state of particle is x₀ ⁱ, initial weight is { Q_o ⁱ=1/n, i= 1,...n}；

(2) state transition model sets up submodule 422: be used for setting up particle state metastasis model, and described particle state turns Shifting formwork type employing following formula:

$x_m^i={\mathrm{Ax}}_{m-1}^i+v_m^i$

In formula,The new particle in expression m moment, m >=2,Being the white Gaussian noise of 0 for average, A is 4 rank unit matrix；m- The particle in 1 moment is propagated by state transition model；

(3) observation model sets up submodule 423, for by color histogram, textural characteristics rectangular histogram and movement edge The mode that feature combines sets up particle observation model；

(4) candidate motion region calculating sub module 424: it utilizes minimum variance estimate to calculate candidate motion region:

$x_{now}=\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_m^j\·x_m^j$

In formula, x_nowRepresent the candidate motion region of the current frame image calculated,Represent the correspondence of m moment jth particle State value；

(5) position correction submodule 425: be used for revising abnormal data:

$x_{pre}=\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_{m-1}^j\·x_{m-1}^j$

In formula, x_preRepresent the candidate motion region of the current frame image calculated,Represent m-1 moment jth particle Corresponding states value；

Data exception evaluation function P=is set | x_now-x_pre|, if the value of P is more than the empirical value T, then x set_now=x_pre；

(6) resampling submodule 426: for deleting, by re-sampling operations, the particle that weights are too small, during resampling, utilize The difference of the prediction of system current time and observation provides and newly ceases residual error, and then carries out the particle of sampling by measuring new breath residual error Online adaptive adjusts, and in sampling process, the contextual definition between number of particles and information residual error is:

Wherein, N_mRepresent the number of particles in m moment, N in sampling process_maxAnd N_minRepresent minimum and maximum particle respectively Number, N_min+1Represent and be merely greater than N_minPopulation, N_max-1Represent and be only smaller than N_maxPopulation,The new breath of etching system when representing m Residual error.

This preferred embodiment uses the side combined based on color histogram, textural characteristics rectangular histogram and motion edge character Formula carries out the right value update of sampling particle, effectively enhances the robustness of tracking system；Position correction submodule 425, energy are set Enough avoid the impact that whole system is brought by abnormal data；In resampling submodule 426, utilize current time prediction and observation Difference provide and newly cease residual error, and then by measuring new breath residual error, the particle of sampling is carried out online adaptive adjustment, and fixed Relation between number of particles and information residual error in justice sampling process, preferably ensure that high efficiency and the algorithm of particle sampler Real-time.

Preferably, the particle right value update formula of described particle observation model is:

$Q_m^j=\frac{\stackrel{\‾}{Q_{Cm}^j}\·\stackrel{\‾}{Q_{Mm}^j}\·\stackrel{\‾}{Q_{Wm}^j}+{\mathrm{\λ}}_1\stackrel{\‾}{Q_{Cm}^j}+{{\mathrm{\λ}}_2}^2\stackrel{\‾}{Q_{Mm}^j}+{{\mathrm{\λ}}_2}^3\stackrel{\‾}{Q_{Wm}^j}+{\mathrm{\λ}}_1{\mathrm{\λ}}_2{\mathrm{\λ}}_3}{(1+{\mathrm{\λ}}_1)(1+{\mathrm{\λ}}_2)(1+{\mathrm{\λ}}_3)}$

In formula

$\stackrel{\‾}{Q_{Cm}^j}=Q_{Cm}^j/\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_{Cm}^j,Q_{Cm}^j=Q_{C(m-1)}^j\frac{1}{\sqrt{2\mathrm{\π}\mathrm{\σ}}}\exp (-\frac{A_m^2}{2{\mathrm{\σ}}^2})$

$\stackrel{\‾}{Q_{Mm}^j}=Q_{Mm}^j/\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_{Mm}^j,Q_{Mm}^j=Q_{M(m-1)}^j\frac{1}{\sqrt{2\mathrm{\π}\mathrm{\σ}}}\exp (-\frac{B_m^2}{2{\mathrm{\σ}}^2})$

$\stackrel{\‾}{Q_{Wm}^j}=Q_{Wm}^j/\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_{Wm}^j,Q_{Wm}^j=Q_{W(m-1)}^j\frac{1}{\sqrt{2\mathrm{\π}\mathrm{\σ}}}\exp (-\frac{C_m^2}{2{\mathrm{\σ}}^2})$

Wherein,Represent the final updated weights of m moment jth particle,WithRepresent m moment and m-1 respectively Jth particle renewal based on color histogram weights in moment,Represent jth in m moment and m-1 moment Particle renewal based on movement edge weights,Represent that in m moment and m-1 moment, jth particle is special based on texture Levy histogrammic renewal weights, A_mFor between the observation based on color histogram of jth particle in the m moment and actual value Bhattacharrya distance, B_mFor between the observation based on movement edge of jth particle in the m moment and actual value Bhattacharrya distance, C_mFor jth particle in the m moment based between the histogrammic observation of textural characteristics and actual value Bhattacharrya distance, σ is Gauss likelihood model variance, λ₁Normalized for feature weight based on color histogram The self-adaptative adjustment factor, λ₂For the normalized self-adaptative adjustment factor of feature weight based on movement edge, λ₃For special based on texture Levy the histogrammic feature weight normalized self-adaptative adjustment factor；

The computing formula of the described self-adaptative adjustment factor is:

${\mathrm{\λ}}_s^m={\mathrm{\ξ}}_{m-1}\·\[-\underset{j=1}{\overset{n}{\mathrm{\Σ}}}\left(p_{m-1}^{s/j}\right){\log }_2^{p_{m-1}^{s/j}}\],s=1,2,3;$

Wherein, during s=1,Represent in the m moment the normalized self-adaptative adjustment of feature weight based on color histogram because of Son,For the eigenvalue based on color histogram in m-1 moment observation probability value under j particle；During s=2,Represent The normalized self-adaptative adjustment factor of feature weight based on movement edge in the m moment,For in the m-1 moment based on motion limit The eigenvalue of edge observation probability value under j particle；During s=3,Represent in the m moment based on the histogrammic spy of textural characteristics Levy the weights normalized self-adaptative adjustment factor,For in the m-1 moment based on the histogrammic eigenvalue of textural characteristics at j grain Observation probability value under Zi；ξ_m-1Represent the locus variance yields of all particles in the m-1 moment.

This preferred embodiment proposes particle right value update formula and the calculating of the self-adaptative adjustment factor of particle observation model Formula, carries out fusion treatment to the feature weight of particle, effectively overcomes the defect that additivity merges and the property taken advantage of fusion exists, enters one Step enhances the robustness of tracking system.

In this application scenarios, choosing population n=60, tracking velocity improves 6.5% relatively, and tracking accuracy carries relatively High by 8.4%.

Application scenarios 4

See Fig. 1, Fig. 2, the Human Movement Tracking System under the complex scene of an embodiment of this application scene, including Human motion video acquisition device 1, image preprocess apparatus 2, shooting adjusting apparatus 3 and follow-up mechanism 4, described human motion regards Frequently harvester 1 is for obtaining the video image comprising human body；Described image preprocess apparatus 2 is for the video image gathered Carry out pretreatment, eliminate the impact of video jitter；Described video image is carried out processing obtaining by described follow-up mechanism 4 follows the tracks of object Present frame position, it was predicted that follow the tracks of object motion direction, according to described tracking object present frame position, and determine area-of-interest, At area-of-interest, described tracking object is tracked；Described shooting adjusting apparatus 3 is used for judging that described tracking object is current Whether frame position is in the central area of current picture, if it is, do not adjust photographic head, if it does not, according to described tracking object The direction of motion adjusts photographic head.

Preferably, described carry out described video image processes acquisition tracking object present frame position, including: to described figure The candidate motion region comprising human body is extracted as carrying out processing；In described candidate motion region, obtain human body target；According to institute State human body target, determine tracking object, obtain and record tracking object present frame position；According to the described current framing bit of tracking object Put the described tracking object motion direction of prediction.

The above embodiment of the present invention adjusts the position of photographic head by selected object of following the tracks of, bonding position prediction, it is achieved flat Sliding tracking effect, the present invention need not any auxiliary locator, do not limited by following the tracks of angle, can follow the tracks of people in all directions Body, robustness is not by external influence, thus solves above-mentioned technical problem.

Preferably, the described video image to gathering carries out pretreatment, including: the first frame figure of selected described video image Picture is reference frame, and reference frame is averagely divided into four regions of non-overlapping copies, and W represents the width of image, and H represents figure image height Degree, four area size are 0.5W × 0.5H, start from image upper left according to being followed successively by region 1,2,3,4 clockwise； At the image center location selection area A that next frame receives₀, A₀Size be chosen to be 0.5W × 0.5H；By A₀According to above-mentioned side Method is divided into four image subblock A that size is 0.25W × 0.25H₁、A₂、A₃、A₄, A₁And A₂For estimating in vertical direction Local motion vector, A₃And A₄For estimating the local motion vector in horizontal direction, make A₁、A₂、A₃、A₄Respectively 1,2,3,4 Search optimal coupling in four regions, thus estimate the global motion vector of video sequence, then carry out inverse motion compensation, Eliminate the impact of video jitter.

Video image is carried out steady as processing by this preferred embodiment, it is to avoid successive image is processed and to cause by video jitter Impact, the efficiency of pretreatment is high.

Preferably, described follow-up mechanism 4 includes that area-of-interest determines module 41, candidate motion region extraction module 42 and Follow the tracks of object-location 43；Described area-of-interest determines that module 41 is emerging for determining sense in a two field picture of video image Interest region D₁And in this, as To Template；Described candidate motion region extraction module 42 is used for setting up particle state transfer and seeing Survey model and based on above-mentioned model, use particle filter predicting candidate moving region；Described tracking object-location 43 is used for Described candidate motion region and described To Template are carried out feature similarity tolerance, recognition and tracking object, and records tracking object Present frame position.

This preferred embodiment constructs the module architectures of follow-up mechanism 4.

Preferably, described candidate motion region extraction module 42 includes:

(1) initialization submodule 421: at described area-of-interest D₁Inside randomly select particle that quantity is n right Each particle carries out initialization process, and after initialization process, the original state of particle is x₀ ⁱ, initial weight is { Q_o ⁱ=1/n, i= 1,...n}；

(2) state transition model sets up submodule 422: be used for setting up particle state metastasis model, and described particle state turns Shifting formwork type employing following formula:

$x_m^i={\mathrm{Ax}}_{m-1}^i+v_m^i$

In formula,The new particle in expression m moment, m >=2,Being the white Gaussian noise of 0 for average, A is 4 rank unit matrix；m- The particle in 1 moment is propagated by state transition model；

(3) observation model sets up submodule 423, for by color histogram, textural characteristics rectangular histogram and movement edge The mode that feature combines sets up particle observation model；

(4) candidate motion region calculating sub module 424: it utilizes minimum variance estimate to calculate candidate motion region:

$x_{now}=\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_m^j\·x_m^j$

In formula, x_nowRepresent the candidate motion region of the current frame image calculated,Represent the correspondence of m moment jth particle State value；

(5) position correction submodule 425: be used for revising abnormal data:

$x_{pre}=\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_{m-1}^j\·x_{m-1}^j$

In formula, x_preRepresent the candidate motion region of the current frame image calculated,Represent m-1 moment jth particle Corresponding states value；

Data exception evaluation function P=is set | x_now-x_pre|, if the value of P is more than the empirical value T, then x set_now=x_pre；

(6) resampling submodule 426: for deleting, by re-sampling operations, the particle that weights are too small, during resampling, utilize The difference of the prediction of system current time and observation provides and newly ceases residual error, and then carries out the particle of sampling by measuring new breath residual error Online adaptive adjusts, and in sampling process, the contextual definition between number of particles and information residual error is:

Wherein, N_mRepresent the number of particles in m moment, N in sampling process_maxAnd N_minRepresent minimum and maximum particle respectively Number, N_min+1Represent and be merely greater than N_minPopulation, N_max-1Represent and be only smaller than N_maxPopulation,The new breath of etching system when representing m Residual error.

This preferred embodiment uses the side combined based on color histogram, textural characteristics rectangular histogram and motion edge character Formula carries out the right value update of sampling particle, effectively enhances the robustness of tracking system；Position correction submodule 425, energy are set Enough avoid the impact that whole system is brought by abnormal data；In resampling submodule 426, utilize current time prediction and observation Difference provide and newly cease residual error, and then by measuring new breath residual error, the particle of sampling is carried out online adaptive adjustment, and fixed Relation between number of particles and information residual error in justice sampling process, preferably ensure that high efficiency and the algorithm of particle sampler Real-time.

Preferably, the particle right value update formula of described particle observation model is:

$Q_m^j=\frac{\stackrel{\‾}{Q_{Cm}^j}\·\stackrel{\‾}{Q_{Mm}^j}\·\stackrel{\‾}{Q_{Wm}^j}+{\mathrm{\λ}}_1\stackrel{\‾}{Q_{Cm}^j}+{{\mathrm{\λ}}_2}^2\stackrel{\‾}{Q_{Mm}^j}+{{\mathrm{\λ}}_2}^3\stackrel{\‾}{Q_{Wm}^j}+{\mathrm{\λ}}_1{\mathrm{\λ}}_2{\mathrm{\λ}}_3}{(1+{\mathrm{\λ}}_1)(1+{\mathrm{\λ}}_2)(1+{\mathrm{\λ}}_3)}$

In formula

$\stackrel{\‾}{Q_{Cm}^j}=Q_{Cm}^j/\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_{Cm}^j,Q_{Cm}^j=Q_{C(m-1)}^j\frac{1}{\sqrt{2\mathrm{\π}\mathrm{\σ}}}\exp (-\frac{A_m^2}{2{\mathrm{\σ}}^2})$

$\stackrel{\‾}{Q_{Mm}^j}=Q_{Mm}^j/\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_{Mm}^j,Q_{Mm}^j=Q_{M(m-1)}^j\frac{1}{\sqrt{2\mathrm{\π}\mathrm{\σ}}}\exp (-\frac{B_m^2}{2{\mathrm{\σ}}^2})$

$\stackrel{\‾}{Q_{Wm}^j}=Q_{Wm}^j/\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_{Wm}^j,Q_{Wm}^j=Q_{W(m-1)}^j\frac{1}{\sqrt{2\mathrm{\π}\mathrm{\σ}}}\exp (-\frac{C_m^2}{2{\mathrm{\σ}}^2})$

Wherein,Represent the final updated weights of m moment jth particle,WithRepresent m moment and m-1 respectively Jth particle renewal based on color histogram weights in moment,Represent jth in m moment and m-1 moment Particle renewal based on movement edge weights,Represent that in m moment and m-1 moment, jth particle is special based on texture Levy histogrammic renewal weights, A_mFor between the observation based on color histogram of jth particle in the m moment and actual value Bhattacharrya distance, B_mFor between the observation based on movement edge of jth particle in the m moment and actual value Bhattacharrya distance, C_mFor jth particle in the m moment based between the histogrammic observation of textural characteristics and actual value Bhattacharrya distance, σ is Gauss likelihood model variance, λ₁Normalized for feature weight based on color histogram The self-adaptative adjustment factor, λ₂For the normalized self-adaptative adjustment factor of feature weight based on movement edge, λ₃For special based on texture Levy the histogrammic feature weight normalized self-adaptative adjustment factor；

The computing formula of the described self-adaptative adjustment factor is:

${\mathrm{\λ}}_s^m={\mathrm{\ξ}}_{m-1}\·\[-\underset{j=1}{\overset{n}{\mathrm{\Σ}}}\left(p_{m-1}^{s/j}\right){\log }_2^{p_{m-1}^{s/j}}\],s=1,2,3;$

Wherein, during s=1,Represent in the m moment the normalized self-adaptative adjustment of feature weight based on color histogram because of Son,For the eigenvalue based on color histogram in m-1 moment observation probability value under j particle；During s=2,Represent The normalized self-adaptative adjustment factor of feature weight based on movement edge in the m moment,For in the m-1 moment based on motion limit The eigenvalue of edge observation probability value under j particle；During s=3,Represent in the m moment based on the histogrammic spy of textural characteristics Levy the weights normalized self-adaptative adjustment factor,For in the m-1 moment based on the histogrammic eigenvalue of textural characteristics at j grain Observation probability value under Zi；ξ_m-1Represent the locus variance yields of all particles in the m-1 moment.

This preferred embodiment proposes particle right value update formula and the calculating of the self-adaptative adjustment factor of particle observation model Formula, carries out fusion treatment to the feature weight of particle, effectively overcomes the defect that additivity merges and the property taken advantage of fusion exists, enters one Step enhances the robustness of tracking system.

In this application scenarios, choosing population n=65, tracking velocity improves 6.5% relatively, and tracking accuracy carries relatively High by 8.5%.

Application scenarios 5

See Fig. 1, Fig. 2, the Human Movement Tracking System under the complex scene of an embodiment of this application scene, including Human motion video acquisition device 1, image preprocess apparatus 2, shooting adjusting apparatus 3 and follow-up mechanism 4, described human motion regards Frequently harvester 1 is for obtaining the video image comprising human body；Described image preprocess apparatus 2 is for the video image gathered Carry out pretreatment, eliminate the impact of video jitter；Described video image is carried out processing obtaining by described follow-up mechanism 4 follows the tracks of object Present frame position, it was predicted that follow the tracks of object motion direction, according to described tracking object present frame position, and determine area-of-interest, At area-of-interest, described tracking object is tracked；Described shooting adjusting apparatus 3 is used for judging that described tracking object is current Whether frame position is in the central area of current picture, if it is, do not adjust photographic head, if it does not, according to described tracking object The direction of motion adjusts photographic head.

Preferably, described carry out described video image processes acquisition tracking object present frame position, including: to described figure The candidate motion region comprising human body is extracted as carrying out processing；In described candidate motion region, obtain human body target；According to institute State human body target, determine tracking object, obtain and record tracking object present frame position；According to the described current framing bit of tracking object Put the described tracking object motion direction of prediction.

The above embodiment of the present invention adjusts the position of photographic head by selected object of following the tracks of, bonding position prediction, it is achieved flat Sliding tracking effect, the present invention need not any auxiliary locator, do not limited by following the tracks of angle, can follow the tracks of people in all directions Body, robustness is not by external influence, thus solves above-mentioned technical problem.

Preferably, the described video image to gathering carries out pretreatment, including: the first frame figure of selected described video image Picture is reference frame, and reference frame is averagely divided into four regions of non-overlapping copies, and W represents the width of image, and H represents figure image height Degree, four area size are 0.5W × 0.5H, start from image upper left according to being followed successively by region 1,2,3,4 clockwise； At the image center location selection area A that next frame receives₀, A₀Size be chosen to be 0.5W × 0.5H；By A₀According to above-mentioned side Method is divided into four image subblock A that size is 0.25W × 0.25H₁、A₂、A₃、A₄, A₁And A₂For estimating in vertical direction Local motion vector, A₃And A₄For estimating the local motion vector in horizontal direction, make A₁、A₂、A₃、A₄Respectively 1,2,3,4 Search optimal coupling in four regions, thus estimate the global motion vector of video sequence, then carry out inverse motion compensation, Eliminate the impact of video jitter.

Video image is carried out steady as processing by this preferred embodiment, it is to avoid successive image is processed and to cause by video jitter Impact, the efficiency of pretreatment is high.

Preferably, described follow-up mechanism 4 includes that area-of-interest determines module 41, candidate motion region extraction module 42 and Follow the tracks of object-location 43；Described area-of-interest determines that module 41 is emerging for determining sense in a two field picture of video image Interest region D₁And in this, as To Template；Described candidate motion region extraction module 42 is used for setting up particle state transfer and seeing Survey model and based on above-mentioned model, use particle filter predicting candidate moving region；Described tracking object-location 43 is used for Described candidate motion region and described To Template are carried out feature similarity tolerance, recognition and tracking object, and records tracking object Present frame position.

This preferred embodiment constructs the module architectures of follow-up mechanism 4.

Preferably, described candidate motion region extraction module 42 includes:

(1) initialization submodule 421: at described area-of-interest D₁Inside randomly select particle that quantity is n right Each particle carries out initialization process, and after initialization process, the original state of particle is x₀ ⁱ, initial weight is { Q_o ⁱ=1/n, i= 1,...n}；

(2) state transition model sets up submodule 422: be used for setting up particle state metastasis model, and described particle state turns Shifting formwork type employing following formula:

$x_m^i={\mathrm{Ax}}_{m-1}^i+v_m^i$

In formula,The new particle in expression m moment, m >=2,Being the white Gaussian noise of 0 for average, A is 4 rank unit matrix；m- The particle in 1 moment is propagated by state transition model；

(3) observation model sets up submodule 423, for by color histogram, textural characteristics rectangular histogram and movement edge The mode that feature combines sets up particle observation model；

(4) candidate motion region calculating sub module 424: it utilizes minimum variance estimate to calculate candidate motion region:

$x_{now}=\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_m^j\·x_m^j$

In formula, x_nowRepresent the candidate motion region of the current frame image calculated,Represent the correspondence of m moment jth particle State value；

(5) position correction submodule 425: be used for revising abnormal data:

$x_{pre}=\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_{m-1}^j\·x_{m-1}^j$

In formula, x_preRepresent the candidate motion region of the current frame image calculated,Represent m-1 moment jth particle Corresponding states value；

Data exception evaluation function P=is set | x_now-x_pre|, if the value of P is more than the empirical value T, then x set_now=x_pre；

(6) resampling submodule 426: for deleting, by re-sampling operations, the particle that weights are too small, during resampling, utilize The difference of the prediction of system current time and observation provides and newly ceases residual error, and then carries out the particle of sampling by measuring new breath residual error Online adaptive adjusts, and in sampling process, the contextual definition between number of particles and information residual error is:

Wherein, N_mRepresent the number of particles in m moment, N in sampling process_maxAnd N_minRepresent minimum and maximum particle respectively Number, N_min+1Represent and be merely greater than N_minPopulation, N_max-1Represent and be only smaller than N_maxPopulation,The new breath of etching system when representing m Residual error.

This preferred embodiment uses the side combined based on color histogram, textural characteristics rectangular histogram and motion edge character Formula carries out the right value update of sampling particle, effectively enhances the robustness of tracking system；Position correction submodule 425, energy are set Enough avoid the impact that whole system is brought by abnormal data；In resampling submodule 426, utilize current time prediction and observation Difference provide and newly cease residual error, and then by measuring new breath residual error, the particle of sampling is carried out online adaptive adjustment, and fixed Relation between number of particles and information residual error in justice sampling process, preferably ensure that high efficiency and the algorithm of particle sampler Real-time.

Preferably, the particle right value update formula of described particle observation model is:

$Q_m^j=\frac{\stackrel{\‾}{Q_{Cm}^j}\·\stackrel{\‾}{Q_{Mm}^j}\·\stackrel{\‾}{Q_{Wm}^j}+{\mathrm{\λ}}_1\stackrel{\‾}{Q_{Cm}^j}+{{\mathrm{\λ}}_2}^2\stackrel{\‾}{Q_{Mm}^j}+{{\mathrm{\λ}}_2}^3\stackrel{\‾}{Q_{Wm}^j}+{\mathrm{\λ}}_1{\mathrm{\λ}}_2{\mathrm{\λ}}_3}{(1+{\mathrm{\λ}}_1)(1+{\mathrm{\λ}}_2)(1+{\mathrm{\λ}}_3)}$

In formula

$\stackrel{\‾}{Q_{Cm}^j}=Q_{Cm}^j/\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_{Cm}^j,Q_{Cm}^j=Q_{C(m-1)}^j\frac{1}{\sqrt{2\mathrm{\π}\mathrm{\σ}}}\exp (-\frac{A_m^2}{2{\mathrm{\σ}}^2})$

$\stackrel{\‾}{Q_{Mm}^j}=Q_{Mm}^j/\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_{Mm}^j,Q_{Mm}^j=Q_{M(m-1)}^j\frac{1}{\sqrt{2\mathrm{\π}\mathrm{\σ}}}\exp (-\frac{B_m^2}{2{\mathrm{\σ}}^2})$

$\stackrel{\‾}{Q_{Wm}^j}=Q_{Wm}^j/\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{Q}_{Wm}^j,Q_{Wm}^j=Q_{W(m-1)}^j\frac{1}{\sqrt{2\mathrm{\π}\mathrm{\σ}}}\exp (-\frac{C_m^2}{2{\mathrm{\σ}}^2})$

Wherein,Represent the final updated weights of m moment jth particle,WithRepresent m moment and m-1 respectively Jth particle renewal based on color histogram weights in moment,Represent jth in m moment and m-1 moment Particle renewal based on movement edge weights,Represent that in m moment and m-1 moment, jth particle is special based on texture Levy histogrammic renewal weights, A_mFor between the observation based on color histogram of jth particle in the m moment and actual value Bhattacharrya distance, B_mFor between the observation based on movement edge of jth particle in the m moment and actual value Bhattacharrya distance, C_mFor jth particle in the m moment based between the histogrammic observation of textural characteristics and actual value Bhattacharrya distance, σ is Gauss likelihood model variance, λ₁Normalized for feature weight based on color histogram The self-adaptative adjustment factor, λ₂For the normalized self-adaptative adjustment factor of feature weight based on movement edge, λ₃For special based on texture Levy the histogrammic feature weight normalized self-adaptative adjustment factor；

The computing formula of the described self-adaptative adjustment factor is:

${\mathrm{\λ}}_s^m={\mathrm{\ξ}}_{m-1}\·\[-\underset{j=1}{\overset{n}{\mathrm{\Σ}}}\left(p_{m-1}^{s/j}\right){\log }_2^{p_{m-1}^{s/j}}\],s=1,2,3;$

Wherein, during s=1,Represent in the m moment the normalized self-adaptative adjustment of feature weight based on color histogram because of Son,For the eigenvalue based on color histogram in m-1 moment observation probability value under j particle；During s=2,Represent The normalized self-adaptative adjustment factor of feature weight based on movement edge in the m moment,For in the m-1 moment based on motion limit The eigenvalue of edge observation probability value under j particle；During s=3,Represent in the m moment based on the histogrammic spy of textural characteristics Levy the weights normalized self-adaptative adjustment factor,For in the m-1 moment based on the histogrammic eigenvalue of textural characteristics at j grain Observation probability value under Zi；ξ_m-1Represent the locus variance yields of all particles in the m-1 moment.

This preferred embodiment proposes particle right value update formula and the calculating of the self-adaptative adjustment factor of particle observation model Formula, carries out fusion treatment to the feature weight of particle, effectively overcomes the defect that additivity merges and the property taken advantage of fusion exists, enters one Step enhances the robustness of tracking system.

In this application scenarios, choosing population n=70, tracking velocity improves 6% relatively, and tracking accuracy improves relatively 9%

Last it should be noted that, use above scene is only in order to illustrate technical scheme, rather than to the present invention The restriction of protection domain, although having made to explain to the present invention with reference to preferred application scene, the ordinary skill people of this area Member should be appreciated that and can modify technical scheme or equivalent, without deviating from technical solution of the present invention Spirit and scope.

Claims (3)

1. a Human Movement Tracking System, it is characterised in that include that human motion video acquisition device, Image semantic classification fill Putting, shoot adjusting apparatus and follow-up mechanism, described human motion video acquisition device is for obtaining the video image comprising human body； Described image preprocess apparatus, for the video image gathered carries out pretreatment, eliminates the impact of video jitter；Described tracking Described video image is carried out processing obtaining by device follows the tracks of object present frame position, it was predicted that follow the tracks of object motion direction, according to institute State tracking object present frame position, and determine area-of-interest, at area-of-interest, described tracking object is tracked；Described Shooting adjusting apparatus for judge described tracking object present frame position whether in the central area of current picture, if it is, Do not adjust photographic head, if it does not, adjust photographic head according to described tracking object motion direction.

A kind of Human Movement Tracking System the most according to claim 1, it is characterised in that described described video image is entered Row processes to obtain follows the tracks of object present frame position, including: carry out described image processing and extract the Candidate Motion district comprising human body Territory；In described candidate motion region, obtain human body target；According to described human body target, determine tracking object, obtain and record Follow the tracks of object present frame position；According to following the tracks of object motion direction described in described tracking object present frame position prediction.

A kind of Human Movement Tracking System the most according to claim 2, it is characterised in that the described video image to gathering Carry out pretreatment, including: the first two field picture of selected described video image is reference frame, and is averagely divided into by reference frame the most not Four overlapping regions, W represents the width of image, and H represents picture altitude, and four area size are 0.5W × 0.5H, from figure As upper left starts according to being followed successively by region 1,2,3,4 clockwise；The image center location received at next frame selectes district Territory A₀, A₀Size be chosen to be 0.5W × 0.5H；By A₀It is divided into four figures that size is 0.25W × 0.25H according to the method described above As sub-block A₁、A₂、A₃、A₄, A₁And A₂For estimating the local motion vector in vertical direction, A₃And A₄For estimating horizontal direction On local motion vector, make A₁、A₂、A₃、A₄In 1,2,3,4 four regions, search optimal coupling respectively, thus estimate and regard The global motion vector of frequency sequence, then carries out inverse motion compensation, eliminates the impact of video jitter.