CN106127808A - A kind of block particle filter method for tracking target based on color and the anti-of local binary patterns Feature Fusion - Google Patents

Abstract

A kind of block particle filter method for tracking target based on color and the anti-of local binary patterns Feature Fusion, including: the initialization of target；The color integration histogram of area-of-interest and local binary patterns integration histogram feature extraction；The feature deterministic coefficient of color characteristic and local binary patterns feature calculates；Different according to current goal state, select different trackings: if dbjective state is normal, target following is carried out with the particle filter method of color Yu local binary patterns Feature Fusion, if dbjective state is partial occlusion, target following is carried out with the piecemeal particle filter method of color Yu local binary patterns Feature Fusion, if dbjective state is for seriously to block, use least square model target location；Update current goal state；When target is in normal condition, more fresh target；The resampling of particle；Particle propagation.The present invention can improve stability and the robustness of target following under circumstance of occlusion.

Description

A kind of block particle filter mesh based on color and the anti-of local binary patterns Feature Fusion Mark tracking

Technical field

The present invention relates to the fields such as image procossing, Video processing, target following, particularly relate to target following based on video Field.

Background technology

In video frequency object tracking technology, occlusion issue is a common difficult point.Target, may in moving process Run into various blocking: target because of the oneself rotated or cause because of action block, that mobile target runs into that other pedestrians cause is mutual Block, mobile target is blocked by the barrier in surrounding.Target runs into when blocking so that carrying of clarification of objective information Take disturbed, cause target characteristic obtain imperfect obtain the most completely less than, finally cause target following inaccurate, even mesh Mark is with losing.

Particle filter be a kind of by Monte Carlo simulation and based on Bayesian Estimation derive algorithm, it utilizes shape In state space, one group of random sample with weight (" particle ") represents the posterior probability density function of state, then utilizes shellfish This method of estimation of leaf is carried out constantly iteration and is updated the new random sample of derivation and new weight, obtains subsequent time shape with this The posterior probability density function of state.Particle filter method is applicable to the Target Tracking Problem under non-linear, non-Gaussian filtering.From Seeing in principle, particle filter tracking algorithm had both had the potentiality processing occlusion issue, had again geneogenous Feature Fusion framework.

Traditional particle filter tracking method is extracted single color histogram and is carried out target following, and color histogram is one Kind of feature based on the overall situation, it is not concerned with the local feature of target, so after running into and blocking, global color feature can not Target is described very accurately.

Summary of the invention

In order to overcome currently existing video target tracking method single features limitation and when blocking table The tracking effect revealed is the best, even target with the problem lost, the present invention proposes a kind of special with local binary patterns based on color Levy the anti-of fusion and block particle filter method for tracking target, the method by color characteristic and local binary patterns feature according to each Feature and the discrimination of background, be determined by property coefficient and carry out additivity fusion, can more effectively describe target characteristic, follow the tracks of Cheng Zhong, carries out real-time judgment and takes corresponding follow-up mechanism for different circumstance of occlusions, thus improving screening circumstance of occlusion The stability of target following and robustness in the case of gear.

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

A kind of block particle filter method for tracking target based on color and the anti-of local binary patterns Feature Fusion, including with Lower step:

Step 1, the initialization of target；

Step 2, the color integration histogram of area-of-interest and local binary patterns integration histogram feature extraction；

Step 3, the feature deterministic coefficient of color characteristic and local binary patterns feature calculates: calculate each particle rectangle The color histogram feature of frame and local binary patterns histogram feature, the color histogram calculating each particle background area is special Seek peace local binary patterns histogram feature, calculate the likelihood of the color characteristic of each particle and the color characteristic of its background area Ratio the discrimination according to the color characteristic of each particle of likelihood ratio calculation with background color feature, calculate according to discrimination The feature deterministic coefficient of the color characteristic of each particle, calculates local binary patterns feature and its background area of each particle Local binary patterns feature likelihood ratio and according to the local binary patterns feature of each particle of likelihood ratio calculation and the back of the body The discrimination of scape local binary patterns feature, determines according to the feature of the local binary patterns feature of the discrimination each particle of calculating Property coefficient；

Step 4, different according to current goal state, select different trackings: if dbjective state is normal, use color Carry out target following with the particle filter method of local binary patterns Feature Fusion, if dbjective state is partial occlusion, use face Color carries out target following with the piecemeal particle filter method of local binary patterns Feature Fusion, if dbjective state is serious screening Gear, uses least square model target location；

Step 5, updates current goal state；

Step 6, when target is in normal condition, the color characteristic template of more fresh target, local binary patterns character modules The color characteristic template of plate and sub-block, local binary patterns feature templates；

Step 7, uses system method for resampling to carry out the resampling of particle；

Step 8, particle propagation: through the particle of resampling, on x, y direction, diffusion obtains new corresponding particle respectively, As the initial distribution of particle in next frame.

Further, in described step 1, the initialization procedure of target is: in the 1st frame, manually target selected by frame, note A height of height of target following frame, a width of width, target's center's point coordinates is (x₁,y₁), the color extracting target area is straight Side's figure and local binary patterns feature the color characteristic template H=(h of initialized target₁,h₂,…,h_n) and local binary patterns Feature templates G=(g₁,g₂,…,g_n) (n=1,2 ..., 32), n is the interval number of feature histogram；By the high decile of target Becoming three parts of horizontal sub-blocks, be designated as sub-block 1,2,3 the most respectively, wide by target is divided into three parts of longitudinal sub-blocks, from a left side to Part on the right side and be not designated as sub-block 4,5,6, extract the color histogram of each sub-block and local binary patterns feature initialized target Sub-block color characteristic template H_i=(h'₁,h'₂,…,h'_n) and sub-block local binary patterns feature templates G_i=(g'₁,g'₂,…, g'_n) (i=1,2 ..., 6；N=1,2 ..., 32), initialize population p, initialize the position (p_x of each particle_j,p_y_j) (j= 1,2 ..., p), initialized target state flag bit Flag is 0, initializes the state flag bit of each sub-blockIt is 0.

Further, in described step 2, the color integration histogram of area-of-interest and local binary patterns integration Nogata Figure characteristic extraction procedure is: reading kth frame image P, area-of-interest refers to cover the minimum rectangle of all particle background areas Region, the background area of particle is point centered by particle position, a width ofA height ofRectangular area subtract Going to " going back to " font region behind target rectangle region, wherein, height is the height of target following frame, and width is target following frame Width, the coordinate of four summits A, B, C, D of area-of-interest is respectively as follows:

$(x_A,y_A)=\left(min\right(p\_x)-\frac{\sqrt{2}}{2}height,min(p\_y)-\frac{\sqrt{2}}{2}width),$

$(x_B,y_B)=\left(min\right(p\_x)-\frac{\sqrt{2}}{2}height,max(p\_y)+\frac{\sqrt{2}}{2}width),$

$(x_C,y_C)=\left(max\right(p\_x)+\frac{\sqrt{2}}{2}height,min(p\_y)-\frac{\sqrt{2}}{2}width),$

$(x_D,y_D)=\left(max\right(p\_x)+\frac{\sqrt{2}}{2}height,max(p\_y)+\frac{\sqrt{2}}{2}width),$

Wherein, (p_x, p_y) is the coordinate of particle, and min () is function of minimizing, and max () is maximizing function, meter Calculate the integration histogram H of color characteristic on rectangular area ABCD interested_in(x y), i.e. calculates from picture point P (x_A,y_A) arrive a some P (x, in rectangular area y) constituted color histogram a little；Calculate local binary patterns on rectangular area ABCD interested The integration histogram G of feature_in(x y), i.e. calculates from picture point P (x_A,y_A) to some P, (x owns in rectangular area y) constituted The local binary patterns rectangular histogram of point.

Further, in described step 3, the calculating of the feature deterministic coefficient of color characteristic and local binary patterns feature Process is: utilize integration histogram p particle is extracted respectively with each particle j (j=1,2 ..., p) centered by a width of Width, color histogram feature HP in the rectangle frame of a height of height_j=(hp₁,hp₂,…,hp_n) and local binary patterns Histogram feature GP_j=(gp₁,gp₂,…,gp_n) (n=1,2 ..., 32), wherein four summit A' of particle j rectangle frame, B', C', D' coordinate is respectively as follows:

$(x_{A^{\′}},y_{A^{\′}})=(p\_x_j+\frac{1}{2}height,p\_y_j+\frac{1}{2}width),$

$(x_{B^{\′}},y_{B^{\′}})=(p\_x_j-\frac{1}{2}height,p\_y_j+\frac{1}{2}width),$

$(x_{C^{\′}},y_{C^{\′}})=(p\_x_j+\frac{1}{2}height,p\_y_j-\frac{1}{2}widk),$

$(x_{D^{\′}},y_{D^{\′}})=(p\_x_j-\frac{1}{2}height,p\_y_j-\frac{1}{2}width),$

Wherein, (p_x_j,p_y_j) (j=1,2 ..., p) being the coordinate of particle j, then the color histogram of particle j rectangle frame is special Levy HP_jWith local binary patterns histogram feature GP_jIt is respectively as follows:

HP_j=H_in(x_A',y_A')-H_in(x_C',y_C'-1)-H_in(x_B'-1,y_B')+H_in(x_D'-1,y_D'-1), GP_j=G_in(x_A', y_A')-G_in(x_C',y_C'-1)-G_in(x_B'-1,y_B')+G_in(x_D'-1,y_D'-1), wherein, H_in(x is y) on rectangular area interested The integration histogram of color characteristic, G_in(x y) is the integration histogram of local binary patterns feature on rectangular area interested；Profit Color histogram feature BG_HP of the background area of particle j is extracted with integration histogram_j=(bh₁,bh₂,…,bh_n) and local Binary pattern histogram feature BG_GP_j=(bg₁,bg₂,…,bg_n) (n=1,2 ..., 32), the background area of particle is with grain Centered by sub-position, point, a width ofA height ofRectangular area deduct target rectangle region after " returning " word Shape region, four outer dead centre E', F', G', H' coordinates of particle background area are respectively as follows:

$(x_{E^{\′}},y_{E^{\′}})=(p\_x_j+\frac{\sqrt{2}}{2}height,p\_y_j+\frac{\sqrt{2}}{2}width),$

$(x_{F^{\′}},y_{F^{\′}})=(p\_x_j-\frac{\sqrt{2}}{2}height,p\_y_j+\frac{\sqrt{2}}{2}width),$

$(x_{G^{\′}},y_{G^{\′}})=(p\_x_j+\frac{\sqrt{2}}{2}height,p\_y_j-\frac{\sqrt{2}}{2}width),$

$(x_{H^{\′}},y_{H^{\′}})=(p\_x_j-\frac{\sqrt{2}}{2}height,p\_y_j-\frac{\sqrt{2}}{2}width),$

Then color histogram feature BG_HP of particle j background area_jWith local binary patterns histogram feature BG_GP_jPoint It is not:

BG_HP_j=H_in(x_E',y_E')-H_in(x_G',y_G'-1)-H_in(x_F'-1,y_F')+H_in(x_H'-1,y_H'-1)-HP_j, BG_GP_j =G_in(x_E',y_E')-G_in(x_G',y_G'-1)-G_in(x_F'-1,y_F')+G_in(x_H'-1,y_H'-1)-GP_j, calculate the color characteristic of particle j Likelihood ratio with the color characteristic of its background areaWherein ε=0.001, if T₀The discrimination threshold value being characterized, then the color characteristic of particle j with the discrimination of background color feature isThe feature deterministic coefficient of the color characteristic of particle j is Calculate the likelihood ratio of the local binary patterns feature of particle j and the local binary patterns feature of its background areaWherein ε=0.001, the local binary mould of particle j Formula feature with the discrimination of background local binary patterns feature isThe local two of particle j The feature deterministic coefficient of binarization mode feature is

Further, in described step 4, the subnormal color of dbjective state and the particle of local binary patterns Feature Fusion Filtered target follow the tracks of process be: p particle is calculated respectively particle j (j=1,2 ..., color characteristic HP p)_j=(hp₁, hp₂,…,hp_n) and color of object feature templates H=(h₁,h₂,…,h_n) (n=1,2 ..., 32) Pasteur's coefficientPasteur's distance isCalculate the color characteristic weight of particle jWherein σ=0.05, is normalized the color characteristic weight of each particleP particle is calculated respectively local binary patterns feature GP of particle j_j=(gp₁,gp₂,…,gp_n) With target local binary patterns feature templates G=(g₁,g₂,…,g_n) Pasteur's coefficientPasteur Distance isCalculate the local binary patterns feature weight of particle j Wherein σ=0.05, is normalized the local binary patterns feature weight of each particleFrom upper The feature deterministic coefficient of the color characteristic that can be calculated particle j in one step is β _ C_j, local binary patterns feature Feature deterministic coefficient is β _ L_j, calculating the weight after each particle color and local binary patterns Feature Fusion isIf (feature deterministic coefficient β _ C_j=β _ L_j=0, then make β _ C_j=β _ L_j=0.5), each particle weights is normalizedEach particle coordinate is added by its weight Power obtains the center point coordinate of present frame (kth frame) target

Or: in described step 4, color when target is at least partially obscured and the particle of local binary patterns Feature Fusion Filtered target follows the tracks of process: occur, according to target in previous frame image, the sub-block state flag bit that circumstance of occlusion detectsP particle is calculated respectively particle j (j=1,2 ..., p) inTime effective son Color histogram feature HP of block_{j_i}With local binary patterns histogram feature GP_{j_i}；Color by each effective sub-block of particle j Feature and corresponding sub-block color characteristic template H_iContrast, calculate Pasteur's coefficient of each effective sub-block iTake the average of effective sub-block similarity as corresponding particle j integral part Similarity, remembers that effective sub-block number is M, then Pasteur's coefficient of particle j isPasteur's distance ForCalculate the color characteristic weight of each particleTo each particle Color characteristic weight is normalizedAgain by the local binary patterns of each effective sub-block of particle j Feature and corresponding sub-block local binary patterns feature templates G_iContrast, calculate Pasteur's coefficient of each effective sub-block iTake the average of effective sub-block similarity as corresponding particle j entirety portion The similarity divided, remembers that effective sub-block number is M, then Pasteur's coefficient of particle j isPasteur Distance isCalculate the local binary patterns feature weight of each particle The local binary patterns feature weight of each particle is normalizedCan from previous step The feature deterministic coefficient of the color characteristic being calculated particle j is β _ C_j, the feature of local binary patterns feature Deterministic coefficient is β _ L_j, calculating the weight after each particle color and local binary patterns Feature Fusion isIf (feature deterministic coefficient β _ C_j=β _ L_j=0, then make β_C_j=β _ L_j=0.5), each particle weights is normalizedEach particle coordinate is weighed by it Heavily weighting obtains the center point coordinate of present frame (kth frame) target

Or be: in described step 4, method of least square target prodiction process when target is seriously blocked is: Target's center point coordinates (x according to the most all frames_t,y_t) (t=1,2 ..., k-1), set up equation below:

$\left\{\begin{array}{c}{x}_t=a_{1}t+b_1\\ y_t=a_{2}t+b_2\end{array}\right.$

Each coefficient a is gone out by solving this Equation for Calculating₁,a₂,b₁,b₂, according to formula x_k=a₁k+b₁, y_k=a₂k+b₂Calculate Obtain the center point coordinate (x of target in kth frame_k,y_k)。

Further, in described step 5, the renewal process of dbjective state is: can be calculated current from preceding step Center point coordinate (the x of frame (kth frame) target_k,y_k), calculate color of object histogram feature H in present frame_acc=(h₁′, h₂′,……,h_n') (n=1,2 ..., 32), note present frame color of object feature and color characteristic template H=(h₁,h₂,…,h_n) (n=1,2 ..., 32) similarity beIf the overall similarity threshold value of target is T₁, when B is big In equal to threshold value T₁Time, illustrate that target is normal condition in the current frame, if now dbjective state flag bit Flag is equal to 0, Then keeping constant, otherwise updating current goal state flag bit is 0, i.e. shows that now target has had been detached from blocking；When B is less than Threshold value T₁Time, illustrate that target is blocked in the current frame, extract coordinates of targets (x_k,y_k) each sub-block i on region (i=1,2 ..., 6) color histogram feature is denoted as H_{acc_i}, calculate each sub-block color characteristic and corresponding sub-block color characteristic template H_iSimilarityIf the color characteristic similarity threshold of sub-block is T₂, then:

${\mathrm{Flag}}_{B^i}=\left\{\begin{array}{c}0,B^i<T_2\\ 1,B^i\&GreaterEqual;T_2\end{array}\right.$

I.e. work as BⁱLess than T₂, this sub-block i is invalid sub-block, remembers sub-block state flag bitIt is 0；Work as BⁱIt is more than or equal to T₂, this sub-block i is effective sub-block, remembers sub-block state flag bitIt is 1, adds up the number M of effective sub-block, according to effectively son Block number judges the serious shielding degree of target:

$\mathrm{Flag}=\left\{\begin{array}{c}1,M>2\\ 2,M\&le;2\end{array}\right.$

I.e. when the number M of effective sub-block is more than 2, illustrate that target is at least partially obscured in the current frame, update dbjective state Flag bit Flag is 1, when the number M of effective sub-block is less than or equal to 2, illustrates that target is seriously blocked in the current frame, updates Dbjective state flag bit Flag is 2.

In described step 6, template renewal method is: note color of object feature templates is H, the target new coordinate district of present frame Territory color histogram is characterized as H_acc, then template renewal formula is: H=α H+ (1-α) H_acc, wherein, 0.80≤α≤0.99, α has Body numerical value sets according to video situation；The local binary patterns feature templates of target, the color characteristic template of sub-block, the office of sub-block Portion's binary pattern feature templates update method is similar with above-mentioned color of object feature templates update method.

Beneficial effects of the present invention is mainly manifested in: extract color characteristic and the local binary patterns feature of target, and will Color characteristic and local binary patterns feature are determined by property coefficient and carry out additivity fusion, can more effectively describe target characteristic, Carry out real-time judgment to blocking and take different follow-up mechanism according to different circumstance of occlusions, can improve under circumstance of occlusion target with The stability of track and robustness.

Accompanying drawing explanation

Fig. 1 be the present invention a kind of based on color and local binary patterns Feature Fusion anti-block particle filter target with Track method flow diagram.

Fig. 2 is method of partition schematic diagram.

Fig. 3 is particle target area and background area schematic diagram.

Fig. 4 is area-of-interest schematic diagram.

Fig. 5 is for seriously blocking schematic diagram.

Fig. 6 is the target following effect of test video, and wherein, (a) is that conventional particle filtered target tracking follows the tracks of knot Really ((a)-1 is the 15th frame, and (a)-2 is the 28th frame, and (a)-3 is the 45th frame, and (a)-4 is the 63rd frame, and (a)-5 is the 92nd frame, (a)- 6 is the 102nd frame, and (a)-7 is the 113rd frame, and (a)-8 is the 142nd frame)；B () is that the one that proposes of the present invention is based on color and local Binary pattern Feature Fusion anti-blocks particle filter method for tracking target to be followed the tracks of result ((b)-1 is the 15th frame, and (b)-2 is the 28 frames, (b)-3 is the 45th frame, and (b)-4 is the 63rd frame, and (b)-5 is the 92nd frame, and (b)-6 is the 102nd frame, and (b)-7 is the 113rd frame, B ()-8 is the 142nd frame).

Detailed description of the invention

The invention will be further described with embodiment below in conjunction with the accompanying drawings.

With reference to Fig. 1～Fig. 6, a kind of based on color and local binary patterns Feature Fusion anti-block particle filter target with Track method, comprises the following steps:

Step 1, the initialization of target；

Step 2, the color integration histogram of area-of-interest and local binary patterns integration histogram feature extraction；

Step 3, the feature deterministic coefficient of color characteristic and local binary patterns feature calculates: calculate each particle rectangle The color histogram feature of frame and local binary patterns histogram feature, the color histogram calculating each particle background area is special Seek peace local binary patterns histogram feature, calculate the likelihood of the color characteristic of each particle and the color characteristic of its background area Ratio the discrimination according to the color characteristic of each particle of likelihood ratio calculation with background color feature, calculate according to discrimination The feature deterministic coefficient of the color characteristic of each particle, calculates local binary patterns feature and its background area of each particle Local binary patterns feature likelihood ratio and according to the local binary patterns feature of each particle of likelihood ratio calculation and the back of the body The discrimination of scape local binary patterns feature, determines according to the feature of the local binary patterns feature of the discrimination each particle of calculating Property coefficient；

Step 4, different according to current goal state, select different trackings: if dbjective state is normal, use color Carry out target following with the particle filter method of local binary patterns Feature Fusion, if dbjective state is partial occlusion, use face Color carries out target following with the piecemeal particle filter method of local binary patterns Feature Fusion, if dbjective state is serious screening Gear, uses least square model target location；

Step 5, updates current goal state；

Step 6, when target is in normal condition, the color characteristic template of more fresh target, local binary patterns character modules The color characteristic template of plate and sub-block, local binary patterns feature templates；

Step 7, uses system method for resampling to carry out the resampling of particle；

Step 8, particle propagation: through the particle of resampling, on x, y direction, diffusion obtains new corresponding particle respectively, As the initial distribution of particle in next frame.

The present embodiment uses one section of video of CAVIAR video library to test, and this video is the MPG form of MPEG2 compression File, resolution is 384 × 288 pixels, and frame speed is 25 frames per second, if population is 300, threshold value T₀=0.7, T₁=0.8, T₂ =0.9, α=0.9.

Concrete implementing procedure includes 8 steps, as it is shown in figure 1, particularly as follows:

(1) object initialization

In the 1st frame, manually target selected by frame, and a height of height of note target following frame, a width of width, in target Heart point coordinates is (x₁,y₁), extract color histogram feature and the local binary patterns histogram feature of target area, initialize The color characteristic template H=(h of target₁,h₂,…,h_n) and local binary patterns feature templates G=(g₁,g₂,…,g_n) (n=1, 2 ..., 32), n is the interval number of feature histogram.As in figure 2 it is shown, the height of target is divided into three parts of horizontal sub-blocks, from upper Being designated as sub-block 1,2,3 respectively to lower, wide by target is divided into three parts of longitudinal sub-blocks, is designated as sub-block 4,5,6 the most respectively, Extract color histogram and the local binary patterns histogram feature of each sub-block, the sub-block color characteristic template of initialized target H_i=(h'₁,h'₂,…,h'_n) and sub-block local binary patterns feature templates G_i=(g'₁,g'₂,…,g'_n) (i=1,2 ..., 6；n =1,2 ..., 32), initialize population p, initialize the position (p_x of each particle_j,p_y_j) (j=1,2 ..., p), initialize mesh Mark state flag bit Flag is 0, initializes the state flag bit of each sub-blockIt is 0.

(2) the color integration histogram of area-of-interest and local binary patterns integration histogram feature extraction

Reading kth frame image P, area-of-interest refers to cover the minimum rectangular area of all particle background areas, such as Fig. 3 Represent is original image and position, particle background area, and in figure, rectangle A'B'C'D' is exactly a width of centered by particle Width, the target area of a height of height, rectangle E'F'G'H' is exactly a width of centered by particleA height ofRegion, " returning " font region that rectangle A'B'C'D' and rectangle E'F'G'H' are constituted is the background area of particle, Target area is identical with background area area coverage, and the number of the total pixel i.e. comprised is identical.Represent such as Fig. 4 is former Beginning image and the position of area-of-interest, rectangle ABCD (i.e. oblique line shading part) i.e. area-of-interest, its four summits in figure The coordinate of A, B, C, D is respectively as follows:

$(x_A,y_A)=\left(min\right(p\_x)-\frac{\sqrt{2}}{2}height,min(p\_y)-\frac{\sqrt{2}}{2}width),$

$(x_B,y_B)=\left(min\right(p\_x)-\frac{\sqrt{2}}{2}height,max(p\_y)+\frac{\sqrt{2}}{2}width),$

$(x_C,y_C)=\left(max\right(p\_x)+\frac{\sqrt{2}}{2}height,min(p\_y)-\frac{\sqrt{2}}{2}width),$

$(x_D,y_D)=\left(max\right(p\_x)+\frac{\sqrt{2}}{2}height,max(p\_y)+\frac{\sqrt{2}}{2}width),$

Wherein, min () is function of minimizing, and max () is maximizing function.

In integration histogram, what the value of each pixel represented is that the initial point in the upper left corner from image is to this pixel In the rectangular area constituted color histogram a little.Calculate the integration of color characteristic on rectangular area ABCD interested straight Side figure H_in(x y), i.e. calculates from picture point P (x_A,y_A) to a some P (x, in rectangular area y) constituted color histogram a little Figure；Calculate the integration histogram G of local binary patterns feature on rectangular area ABCD interested_in(x y), i.e. calculates from picture point P(x_A,y_A) to a some P (x, in rectangular area y) constituted local binary patterns rectangular histogram a little.

(3) feature deterministic coefficient calculates

For different tracking environmentals, feature is the most different to the power of the descriptive power of its target.Calculating mesh When marking the feature in region, it should in view of entirety and the differentiation degree of background of target, when target characteristic with background subtraction away from comparing Time big, illustrate that this feature can preferably describe target.The present invention by carrying out the differentiation journey of comparison object and background with log likelihood ratio Degree, thus calculate the deterministic coefficient of feature, revise the particle weights under different characteristic statement by deterministic coefficient, obtain One particle weights more accurately.

Utilize the integration histogram H of the color characteristic of area-of-interest_in(x, y) straight with the integration of local binary patterns feature Side figure G_in(x y), can quickly be calculated a width of width centered by each particle j, a height of height in current kth frame Region in color histogram feature HP_j=(hp₁,hp₂,…,hp_n) and local binary patterns histogram feature GP_j=(gp₁, gp₂,…,gp_n) (n=1,2 ..., 32).As it is shown on figure 3, the target area of particle j is rectangle A'B'C'D', its four summits The coordinate of A', B', C', D' is respectively as follows:

$(x_{A^{\&prime;}},y_{A^{\&prime;}})=(p\_x_j+\frac{1}{2}height,p\_y_j+\frac{1}{2}width),$

$(x_{B^{\&prime;}},y_{B^{\&prime;}})=(p\_x_j-\frac{1}{2}height,p\_y_j+\frac{1}{2}width),$

$(x_{C^{\&prime;}},y_{C^{\&prime;}})=(p\_x_j+\frac{1}{2}height,p\_y_j-\frac{1}{2}width),$

$(x_{D^{\&prime;}},y_{D^{\&prime;}})=(p\_x_j-\frac{1}{2}height,p\_y_j-\frac{1}{2}width),$

Then utilize integration histogram can calculate color histogram feature HP of particle j rectangle frame_jStraight with local binary patterns Side's figure feature GP_jIt is respectively as follows:

HP_j=H_in(x_A',y_A')-H_in(x_C',y_C'-1)-H_in(x_B'-1,y_B')+H_in(x_D'-1,y_D'-1) (1)

GP_j=G_in(x_A',y_A')-G_in(x_C',y_C'-1)-G_in(x_B'-1,y_B')+G_in(x_D'-1,y_D'-1) (2)

The background area that " returning " font region is particle j that rectangle A'B'C'D' and rectangle E'F'G'H' is constituted, background area The coordinate of four outer dead centres E', F', G', the H' in territory is respectively as follows:

$(x_{E^{\&prime;}},y_{E^{\&prime;}})=(p\_x_j+\frac{\sqrt{2}}{2}height,p\_y_j+\frac{\sqrt{2}}{2}width),$

$(x_{F^{\&prime;}},y_{F^{\&prime;}})=(p\_x_j-\frac{\sqrt{2}}{2}height,p\_y_j+\frac{\sqrt{2}}{2}width),$

$(x_{G^{\&prime;}},y_{G^{\&prime;}})=(p\_x_j+\frac{\sqrt{2}}{2}height,p\_y_j-\frac{\sqrt{2}}{2}width),$

$(x_{H^{\&prime;}},y_{H^{\&prime;}})=(p\_x_j-\frac{\sqrt{2}}{2}height,p\_y_j-\frac{\sqrt{2}}{2}width),$

Integration histogram is then utilized to extract the color histogram BG_HP of particle j background area_jWith local binary patterns Nogata Figure BG_GP_jIt is respectively as follows:

BG_HP_j=H_in(x_E',y_E')-H_in(x_G',y_G'-1)-H_in(x_F'-1,y_F')+H_in(x_H'-1,y_H'-1)-HP_j (3)

BG_GP_j=G_in(x_E',y_E')-G_in(x_G',y_G'-1)-G_in(x_F'-1,y_F')+G_in(x_H'-1,y_H'-1)-GP_j (4)

By calculating log likelihood ratio function, can comparison object feature histogram and background characteristics rectangular histogram more quickly, Pick out the obvious characteristic component of discrimination.Calculate the likelihood of the color characteristic of particle j and the color characteristic of its background area Ratio:

$LR\_C_j\left(n\right)=\left|\log \frac{max\{{\mathrm{HP}}_j\left(n\right),\mathrm{\&epsiv;}\}}{max\{BG\_{\mathrm{HP}}_j\left(n\right),\mathrm{\&epsiv;}\}}\right|---\left(5\right)$

Wherein, ε=0.001, it is ensured that molecule denominator can not be all zero.If T₀The discrimination threshold value that is characterized, then particle j Color characteristic with the discrimination of background color feature is:

$\mathrm{\&gamma;}\_c_j\left(n\right)=\left\{\begin{array}{c}1,LR\_C_j\left(n\right)\&GreaterEqual;T_0\\ 0,LR\_C_j\left(n\right)<T_0\end{array}\right.---\left(6\right)$

In the present invention, the deterministic coefficient of target characteristic refers to that feature bigger with background characteristics gap in target characteristic accounts for The ratio of whole target total characteristic, then the feature deterministic coefficient of color characteristic is:

$\mathrm{\&beta;}\_C_j=\frac{\underset{q=1}{\overset{n}{\&Sigma;}}\mathrm{\&gamma;}\_c_j\left(q\right){\mathrm{HP}}_j\left(q\right)}{\underset{q=1}{\overset{n}{\&Sigma;}}{\mathrm{HP}}_j\left(q\right)}---\left(7\right)$

The local binary patterns feature of calculating particle j and the likelihood ratio of the local binary patterns feature of its background area:

$LR\_L_j\left(n\right)=\left|\log \frac{max\{{\mathrm{GP}}_j\left(n\right),\mathrm{\&epsiv;}\}}{max\{BG\_{\mathrm{GP}}_j\left(n\right),\mathrm{\&epsiv;}\}}\right|---\left(8\right)$

The then local binary patterns feature of particle j and the discrimination of background local binary patterns feature:

$\mathrm{\&gamma;}\_l_j\left(n\right)=\left\{\begin{array}{c}1,LR\_L_j\left(n\right)\&GreaterEqual;T_0\\ 0,LR\_L_j\left(n\right)<T_0\end{array}\right.---\left(9\right)$

The feature deterministic coefficient of its local binary patterns feature is:

$\mathrm{\&beta;}\_L_j=\frac{\underset{q=1}{\overset{n}{\&Sigma;}}\mathrm{\&gamma;}\_l_j\left(q\right){\mathrm{GP}}_j\left(q\right)}{\underset{q=1}{\overset{n}{\&Sigma;}}{\mathrm{GP}}_j\left(q\right)}---\left(10\right)$

(4) selection of tracking strategy

According to the difference of dbjective state, select different tracking strategy to reach anti-and block and the purpose of tenacious tracking.As Really dbjective state flag bit Flag is 0, represents that dbjective state is normal, with the particle filter of color with local binary patterns Feature Fusion Wave method carries out target following；If dbjective state flag bit Flag is 1, expression dbjective state is partial occlusion, with color with The piecemeal particle filter method of local binary patterns Feature Fusion carries out target following；If dbjective state flag bit Flag is 2, Represent that dbjective state, for seriously to block, uses least square model target location.

The subnormal color of dbjective state with the particle filter object tracking process of local binary patterns Feature Fusion is: root Color histogram feature HP according to particle j rectangle frame calculated in previous step_jWith local binary patterns histogram feature GP_j, with Pasteur's coefficient as current particle frame feature and the reference coefficient of To Template similarity contrast, when the number of Pasteur's coefficient Being worth the biggest, the numerical value of Pasteur's distance is the least, illustrates that the similarity degree of two samples is the highest, otherwise, illustrate between two samples Similarity degree the lowest.The color characteristic HP of particle j_jPasteur's coefficient B C with color of object feature templates H_jComputing formula is:

${\mathrm{BC}}_j=\underset{q=1}{\overset{n}{\&Sigma;}}\sqrt{{\mathrm{HP}}_j\left(q\right)H\left(q\right)}---\left(11\right)$

Pasteur's distance isColor characteristic weight w of particle j₁J () computing formula is:

$w_1\left(j\right)=\frac{1}{\sqrt{2\mathrm{\&pi;}\mathrm{\&sigma;}}}\exp (-\frac{{{\mathrm{dC}}_j}^2}{2{\mathrm{\&sigma;}}^2})---\left(12\right)$

Wherein, σ=0.05, more each particle color feature weight is normalized:

$w_1\left(j\right)=\frac{w_1\left(j\right)}{\underset{q=1}{\overset{p}{\&Sigma;}}{w}_1\left(q\right)}---\left(13\right)$

Local binary patterns feature GP of particle j_jPasteur's coefficient B L with target local binary patterns feature templates G_jCalculate Formula is:

${\mathrm{BL}}_j=\underset{q=1}{\overset{n}{\&Sigma;}}\sqrt{{\mathrm{GP}}_j\left(q\right)G\left(q\right)}---\left(14\right)$

Pasteur's distance isThe local binary patterns feature weight w of particle j₂J () computing formula is:

$w_2\left(j\right)=\frac{1}{\sqrt{2\mathrm{\&pi;}\mathrm{\&sigma;}}}\exp (-\frac{{{\mathrm{dL}}_j}^2}{2{\mathrm{\&sigma;}}^2})---\left(15\right)$

Again each particle local binary patterns feature weight is normalized:

$w_2\left(j\right)=\frac{w_2\left(j\right)}{\underset{q=1}{\overset{p}{\&Sigma;}}{w}_2\left(q\right)}---\left(16\right)$

According to feature deterministic coefficient calculated in previous step, color and local binary patterns feature are added Property merge and calculate the weight of each particle:

$w\left(j\right)=\frac{\mathrm{\&beta;}\_C_j}{\mathrm{\&beta;}\_C_j+\mathrm{\&beta;}\_L_j}{w}_1\left(j\right)+\frac{\mathrm{\&beta;}\_L_j}{\mathrm{\&beta;}\_C_j+\mathrm{\&beta;}\_L_j}{w}_2\left(j\right)---\left(17\right)$

Wherein, in order to ensure the effectiveness of formula (17), if feature deterministic coefficient β _ C_jWith β _ L_jBe simultaneously 0 season β _ C_j=β _ L_j=0.5.Each particle weights is normalizedThe coordinate of all particles is weighed by it Heavily weighting obtains the center point coordinate of present frame target

When occurring blocking, target is carried out piecemeal tracking.The block feature being blocked can not complete extraction, now need The sub-block not being blocked is carried out feature extraction, to reach the purpose persistently followed the tracks of.

Color when target is at least partially obscured with the particle filter object tracking process of local binary patterns Feature Fusion is: The sub-block state flag bit according to target in previous frame image, circumstance of occlusion occurring and detectExtract in particle jTime the most effective sub-block color histogram feature HP_{j_i}With local binary patterns histogram feature GP_{j_i}.By particle Color histogram feature HP of each effective sub-block in j rectangle frame_{j_i}With corresponding sub-block color template H_iCarry out comparing calculation each Pasteur's coefficient of sub-block color characteristic:

${\mathrm{BC}}_{j\_i}=\underset{q=1}{\overset{n}{\&Sigma;}}\sqrt{{\mathrm{HP}}_{j\_i}\left(q\right)H_i\left(q\right)}---\left(18\right)$

Take effective sub-block similarity average similarity as the integral part of corresponding particle, remember that effective sub-block number is M, Then Pasteur's coefficient of particle j color characteristic is:

${\mathrm{BC}}_j=\frac{1}{M}\underset{i=1}{\overset{6}{\&Sigma;}}{\mathrm{Flag}}_{B^i}{\mathrm{BC}}_{j\_i}---\left(19\right)$

Pasteur's distance isColor characteristic weight w of each particle is calculated with formula (12)₁J (), by public affairs Each particle color feature weight is normalized by formula (13).

By the local binary patterns histogram feature GP of each effective sub-block in particle j rectangle frame_{j_i}With corresponding sub-block office Portion's binary pattern template G_iCarry out Pasteur's coefficient of comparing calculation each sub-block local binary patterns feature:

${\mathrm{BL}}_{j\_i}=\underset{q=1}{\overset{n}{\&Sigma;}}\sqrt{{\mathrm{GP}}_{j\_i}\left(q\right)G_i\left(q\right)}---\left(20\right)$

Then Pasteur's coefficient of the local binary patterns feature of particle j is:

${\mathrm{BL}}_j=\frac{1}{M}\underset{i=1}{\overset{6}{\&Sigma;}}{\mathrm{Flag}}_{B^i}{\mathrm{BL}}_{j\_i}---\left(21\right)$

Pasteur's distance isThe local binary patterns feature weight w of each particle is calculated with formula (15)₂ J (), is normalized each particle local binary patterns feature weight with formula (16).

Each particle is calculated according to feature deterministic coefficient Feature Fusion formula (17) calculated in previous step Weight w (j), is normalized each particle weightsThe coordinate of all particles is added by its weight Power obtains the center point coordinate of present frame target

Method of least square target prodiction process when target is seriously blocked is: during following the tracks of, if previous Frame detects that target is seriously blocked, then feature information extraction is the most difficult, needs the target location according to above k-1 frame pre- Measure the position of kth frame target.Before note, in frame, the center position coordinates of target is (x_t,y_t), wherein t=1,2 ..., k-1, t Represent frame number.When target was seriously blocked within the of short duration time, it is assumed that target approximation is moving along a straight line.According to least square Method principle, sets up current goal center coordinate x in x-axis_tCoordinate y on the y axis_tEquation of change along with frame number t As follows:

$\left\{\begin{array}{c}{x}_t=a_{1}t+b_1\\ y_t=a_{2}t+b_2\end{array}\right.---\left(22\right)$

Substitute into known center position coordinates (x_t,y_t) (t=1,2 ..., k-1) solve this equation, two of which straight line Slope a₁、a₂With intercept b₁、b₂Computing formula respectively as follows:

$a_1=\frac{(k-1)\&CenterDot;\underset{t=1}{\overset{k-1}{\&Sigma;}}t\&CenterDot;x_t-\underset{t=1}{\overset{k-1}{\&Sigma;}}t\&CenterDot;\underset{t=1}{\overset{k-1}{\&Sigma;}}{x}_t}{(k-1)\&CenterDot;\underset{t=1}{\overset{k-1}{\&Sigma;}}{t}^2-{\left(\underset{t=1}{\overset{k-1}{\&Sigma;}}t\right)}^2}---\left(23\right)$

$a_2=\frac{(k-1)\&CenterDot;\underset{t=1}{\overset{k-1}{\&Sigma;}}t\&CenterDot;y_t-\underset{t=1}{\overset{k-1}{\&Sigma;}}t\&CenterDot;\underset{t=1}{\overset{k-1}{\&Sigma;}}{y}_t}{(k-1)\&CenterDot;\underset{t=1}{\overset{k-1}{\&Sigma;}}{t}^2-{\left(\underset{t=1}{\overset{k-1}{\&Sigma;}}t\right)}^2}---\left(24\right)$

$b_1=\frac{\underset{t=1}{\overset{k-1}{\&Sigma;}}{\left(t\right)}^2\&CenterDot;\underset{t=1}{\overset{k-1}{\&Sigma;}}{x}_t-\underset{t=1}{\overset{k-1}{\&Sigma;}}t\&CenterDot;\underset{t=1}{\overset{k-1}{\&Sigma;}}t\&CenterDot;x_t}{(k-1)\&CenterDot;\underset{t=1}{\overset{k-1}{\&Sigma;}}{t}^2-{\left(\underset{t=1}{\overset{k-1}{\&Sigma;}}t\right)}^2}---\left(25\right)$

$b_2=\frac{\underset{t=1}{\overset{k-1}{\&Sigma;}}{\left(t\right)}^2\&CenterDot;\underset{t=1}{\overset{k-1}{\&Sigma;}}{y}_t-\underset{t=1}{\overset{k-1}{\&Sigma;}}t\&CenterDot;\underset{t=1}{\overset{k-1}{\&Sigma;}}t\&CenterDot;y_t}{(k-1)\&CenterDot;\underset{t=1}{\overset{k-1}{\&Sigma;}}{t}^2-{\left(\underset{t=1}{\overset{k-1}{\&Sigma;}}t\right)}^2}---\left(26\right)$

After trying to achieve the straight line of two matchings, target's center's point position (x of prediction in current frame image_k,y_k) it is expressed as:

x_k=a₁k+b₁ (27)

y_k=a₂k+b₂ (28)

Thus can be calculated the target's center position (x of k frame_k,y_k)。

(6) dbjective state is updated

Target is in motor process, and the global feature extracted when the most not blocking generation should be one Change in the range of Ding or keep constant, but after running into and blocking, the feature of the part that is blocked can change, thus affects whole Body clarification of objective.To integral color feature and color template, we are by relatively analyzing whether target is blocked.

Through abovementioned steps at the centre coordinate (x obtaining present frame target state estimator position_k,y_kAfter), will be with current new coordinate (x_k,y_kCentered by), the color characteristic in the region of point compares with target color template, when similarity is higher than threshold value, says Target on bright new coordinate position is high with the target similarity-rough set of tracking, is coupling, continues to use particle filter tracking method It is tracked；When similarity is less than threshold value, clarification of objective generation large change is described, it is believed that target occurs in that blocks feelings Condition, but concrete serious shielding degree also needs to differentiate further.

Calculate color of object feature histogram H in present frame_acc=(h₁′,h₂′,…,h_n'), note present frame target characteristic with The similarity of color template H is:

$B=\underset{q=1}{\overset{n}{\&Sigma;}}\sqrt{H_{acc}\left(q\right)H\left(q\right)}---\left(29\right)$

If the overall similarity threshold value of target is T₁。

As B >=T₁Time, illustrate that target state in the current frame is normal；If now dbjective state flag bit Flag is equal to 0, Without updating, otherwise updating dbjective state flag bit Flag is 0, illustrates that now target has had been detached from blocking.

As B ＜ T₁Time, illustrate that target occurs in that circumstance of occlusion in the current frame, utilize current goal sub-block color characteristic with The seriousness of piecemeal color template comparative analysis target occlusion situation.By each sub-block color characteristic of current location respectively with target In color template, corresponding block feature compares, if similarity is higher, illustrates that this sub-block state is normal；If sub-block color characteristic With when there is larger difference between corresponding sub-block color template, illustrate to occur in that circumstance of occlusion in this sub-block.Will be with current new seat Mark (x_k,y_kCentered by) point rectangular area piecemeal after, extract each sub-block i (i=1,2 ..., 6) color characteristic histogram note Make H_{acc_i}, each sub-block color characteristic respectively with corresponding sub block feature H in target color template_iComparing, corresponding is similar Degree is designated as Bⁱ:

$B^i=\underset{q=1}{\overset{n}{\&Sigma;}}\sqrt{H_{acc\_i}\left(q\right)H_i\left(q\right)}---\left(30\right)$

If the similarity threshold of molecule block is T₂, add up the similarity situation of each sub-block, set effective sub-block and invalid Sub-block, the state flag bit of note sub-block is Flag_Bi, then:

${\mathrm{Flag}}_{B^i}=\left\{\begin{array}{c}0,B^i<T_2\\ 1,B^i\&GreaterEqual;T_2\end{array}\right.---\left(31\right)$

I.e. work as Bⁱ＜ T₂, orderRepresenting sub-block i is invalid sub-block；Work as Bⁱ≥T₂, orderRepresent sub-block i It it is effective sub-block.If piecemeal of the present invention there being 4 sub-blocks circumstance of occlusion all occurs, then block in the gross area covers target frame big Partial pixel.Thus, the threshold value of invalid sub-block number is set to 4, then the threshold value of effective sub-block number is 2, as it is shown in figure 5, work as When sub-block 1,3,4 and 5 is all invalid sub-block, target is in serious occlusion state.Add up the effective sub-block number in all sub-blocks, The serious shielding degree of target is judged according to effective sub-block number:

$Flag=\left\{\begin{array}{c}1,M>2\\ 2,M\&le;2\end{array}\right.---\left(32\right)$

I.e. as M ＞ 2, represent target and be in partial occlusion state, update dbjective state flag bit Flag=1；When M≤2 Time, represent target and be in serious occlusion state, update dbjective state flag bit Flag=2.

(7) renewal of template

Clarification of objective template initialize be that manual frame is selected during target calculated in video initial frame, along with time Between passage, target, it may happen that change more or less, needs clarification of objective template is carried out adaptive updates.Work as mesh When marking under occlusion, template is not updated, in order to avoid being disturbed by shelter.Only it is in normal condition in target Time lower, To Template is updated.I.e. during Flag=0, by template renewal formula respectively to clarification of objective template and sub-block Feature templates be updated simultaneously.The rectangular histogram making color of object feature templates is H, the target new coordinate (x of present frame_k,y_k) Region color feature rectangular histogram is H_acc, color characteristic template renewal formula is:

H=α H+ (1-α) H_acc (33)

Wherein, 0.80≤α≤0.99, α=0.9 in the present embodiment；The local binary patterns feature templates of target, sub-block Color characteristic template, local binary patterns feature templates update method and the color of object feature templates update method class of sub-block With.

(8) particle resampling

We carry out the resampling of particle with system resampling methods, remove the particle that weights are little, retain or reproduction right It is worth high particle.

(9) propagation of particle

Through the particle of resampling, on x, y direction, diffusion obtains new corresponding particle, as particle in next frame respectively Initial distribution.

Fig. 6 is that the conventional particle filtered target tracking one with present invention proposition is based on color and local binary patterns The anti-of Feature Fusion blocks particle filter method for tracking target tracking effect comparison diagram on test video.At this video field Jing Zhong, the man of black jacket is the target person followed the tracks of, and target is blocked, when adopting by the woman of white jacket in moving process When being tracked with traditional particle filter method for tracking target, tracking box can deviate target location, even leave screening in target After block material, tracking box occurs with wrong phenomenon.And when using the method for the present invention to be tracked, target person run into block with And leave after blocking whole during, tracking box can be accurately tracked by target.The 15th, 28,45 frame time, target be in by Occlusion state, both of which can outline target person, and the method for the present invention can outline target person more accurately；The 63rd During frame, target is blocked by obstructions body, and the tracking box of two kinds of methods still can outline the visible part of target；From 102nd frame starts, and the tracking box of conventional particle filtering method has been partial to shelter, is in wrong with state, until target with Lose, and the method for the present invention can preferably continue to be tracked target person, and tracking box shows relatively accurately, reaches good Tenacious tracking effect.

In order to preferably compare two kinds of methods, use the tracking error of center of target weigh two kinds of methods with Track effect.Tracking error Euclidean distance calculates, and is shown below:

$d=\sqrt{{(x^{\&prime;}-x)}^2+{(y^{\&prime;}-y)}^2}---\left(34\right)$

Wherein, (x', y') represents target's center's point position that tracking records, and (x y) is the reality of every frame in video Target's center's point position, we manually measure acquisition here.Calculate the central point obtained with two kinds of trackings respectively Coordinate and tracking error, be analyzed, and comparative result is as shown in Table 1 and Table 2.It is apparent that either with or without screening Gear impact, what the present invention proposed blocks particle filter method for tracking target based on color and the anti-of local binary patterns Feature Fusion All little than the tracking error of conventional particle filtering method, especially with the increase of frame number, run in target and block and leave screening After block material, traditional particle filter method is with losing target, and the error of the inventive method is always maintained within the specific limits, Show the stable and tracking effect of robust.

Table 1

Table 2

It is clear that on the premise of without departing from true spirit and scope of the present invention, invention described herein is permissible There are many changes.Therefore, all changes that it will be apparent to those skilled in the art that, it is intended to be included in present claims Within the scope of book is contained.Scope of the present invention is only defined by described claims.

Claims (9)

1. blocking a particle filter method for tracking target based on color and the anti-of local binary patterns Feature Fusion, its feature exists In: described method for tracking target comprises the following steps:

Step 1, the initialization of target；

Step 2, the color integration histogram of area-of-interest and local binary patterns integration histogram feature extraction；

Step 3, the feature deterministic coefficient of color characteristic and local binary patterns feature calculates: calculate each particle rectangle frame Color histogram feature and local binary patterns histogram feature, calculate each particle background area color histogram feature and Local binary patterns histogram feature, calculates the likelihood ratio of the color characteristic of each particle and the color characteristic of its background area And according to the discrimination of the color characteristic of each particle of likelihood ratio calculation with background color feature, calculate each according to discrimination The feature deterministic coefficient of the color characteristic of particle, calculates the local binary patterns feature of each particle and the office of its background area The likelihood ratio of portion's binary pattern feature according to the local binary patterns feature of each particle of likelihood ratio calculation and background office The discrimination of portion's binary pattern feature, calculates the feature definitiveness system of the local binary patterns feature of each particle according to discrimination Number；

Step 4, different according to current goal state, select different trackings: if dbjective state is normal, with color and office The particle filter method of portion's binary pattern Feature Fusion carries out target following, if dbjective state is partial occlusion, with color with The piecemeal particle filter method of local binary patterns Feature Fusion carries out target following, if dbjective state is for seriously to block, uses Least square model target location；

Step 5, updates current goal state；

Step 6, when target is in normal condition, the color characteristic template of more fresh target, local binary patterns feature templates with And the color characteristic template of sub-block, local binary patterns feature templates；

Step 7, uses system method for resampling to carry out the resampling of particle；

Step 8, particle propagation: through the particle of resampling, on x, y direction, diffusion obtains new corresponding particle respectively, as The initial distribution of particle in next frame.

A kind of block particle filter target based on color and the anti-of local binary patterns Feature Fusion Tracking, it is characterised in that: in described step 1, the initialization procedure of target is: in the 1st frame, manually mesh selected by frame Mark, a height of height, a width of width of note target following frame, target's center's point coordinates is (x₁,y₁), extract the face of target area The color characteristic template H=(h of Color Histogram and local binary patterns feature initialized target₁,h₂,…,h_n) and local binary Pattern feature template G=(g₁,g₂,…,g_n) (n=1,2 ..., 32), n is the interval number of feature histogram；By the height of target Being divided into three parts of horizontal sub-blocks, be designated as sub-block 1,2,3 the most respectively, wide by target is divided into three parts of longitudinal sub-blocks, from Left-to-right it is designated as sub-block 4,5,6 respectively, extracts the color histogram of each sub-block and local binary patterns feature and initialize mesh Target sub-block color characteristic template H_i=(h'₁,h'₂,…,h'_n) and sub-block local binary patterns feature templates G_i=(g'₁,g '₂,…,g'_n) (i=1,2 ..., 6；N=1,2 ..., 32), initialize population p, initialize the position (p_x of each particle_j,p_ y_j) (j=1,2 ..., p), initialized target state flag bit Flag is 0, initializes the state flag bit of each sub-blockFor 0。

A kind of block particle filter based on color and the anti-of local binary patterns Feature Fusion Method for tracking target, it is characterised in that: in described step 2, the color integration histogram of area-of-interest and local binary patterns Integration histogram characteristic extraction procedure is: reading kth frame image P, area-of-interest refers to cover all particle background areas Minimum rectangular area, the background area of particle is point centered by particle position, a width ofA height ofSquare Shape region deduct target rectangle region after " returning " font region, wherein, height is the height of target following frame, and width is mesh The width of mark tracking box, the coordinate of four summits A, B, C, D of area-of-interest is respectively as follows:

$(x_A,y_A)=\left(min\right(p\_x)-\frac{\sqrt{2}}{2}height,min(p\_y)-\frac{\sqrt{2}}{2}width),$

$(x_B,y_B)=\left(min\right(p\_x)-\frac{\sqrt{2}}{2}height,max(p\_y)+\frac{\sqrt{2}}{2}width),$

$(x_C,y_C)=\left(max\right(p\_x)+\frac{\sqrt{2}}{2}height,min(p\_y)-\frac{\sqrt{2}}{2}width),$

$(x_D,y_D)=\left(max\right(p\_x)+\frac{\sqrt{2}}{2}height,max(p\_y)+\frac{\sqrt{2}}{2}width),$

Wherein, (p_x, p_y) is the coordinate of particle, and min () is function of minimizing, and max () is maximizing function, calculates sense The integration histogram H of color characteristic on the ABCD of interest rectangular area_in(x y), i.e. calculates from picture point P (x_A,y_A) to some a P (x, Y) in the rectangular area constituted color histogram a little；Calculate local binary patterns feature on rectangular area ABCD interested Integration histogram G_in(x y), i.e. calculates from picture point P (x_A,y_A) to a some P (x, in rectangular area y) constituted a little Local binary patterns rectangular histogram.

A kind of block particle filter based on color and the anti-of local binary patterns Feature Fusion Method for tracking target, it is characterised in that: in described step 3, the feature definitiveness system of color characteristic and local binary patterns feature Number calculating process be: utilize integration histogram p particle is extracted respectively with each particle j (j=1,2 ..., p) centered by A width of width, color histogram feature HP in the rectangle frame of a height of height_j=(hp₁,hp₂,…,hp_n) and local two Binarization mode histogram feature GP_j=(gp₁,gp₂,…,gp_n) (n=1,2 ..., 32), wherein four summits of particle j rectangle frame A', B', C', D' coordinate is respectively as follows:

$(x_{A^{\&prime;}},y_{A^{\&prime;}})=(p\_x_j+\frac{1}{2}height,p\_y_j+\frac{1}{2}width),$

$(x_{B^{\&prime;}},y_{B^{\&prime;}})=(p\_x_j-\frac{1}{2}height,p\_y_j+\frac{1}{2}width),$

$(x_{C^{\&prime;}},y_{C^{\&prime;}})=(p\_x_j+\frac{1}{2}height,p\_y_j-\frac{1}{2}width),$

$(x_{D^{\&prime;}},y_{D^{\&prime;}})=(p\_x_j-\frac{1}{2}height,p\_y_j-\frac{1}{2}width),$

Wherein, (p_x_j,p_y_j) (j=1,2 ..., p) it is the coordinate of particle j, then the color histogram feature of particle j rectangle frame HP_jWith local binary patterns histogram feature GP_jIt is respectively as follows:

HP_j=H_in(x_A',y_A')-H_in(x_C',y_C'-1)-H_in(x_B'-1,y_B')+H_in(x_D'-1,y_D'-1),

GP_j=G_in(x_A',y_A')-G_in(x_C',y_C'-1)-G_in(x_B'-1,y_B')+G_in(x_D'-1,y_D'-1),

Wherein, H_in(x y) is the integration histogram of color characteristic, G on rectangular area interested_in(x y) is rectangle region interested The integration histogram of local binary patterns feature on territory；Utilize the color histogram of the background area of integration histogram extraction particle j Figure feature BG_HP_j=(bh₁,bh₂,…,bh_n) and local binary patterns histogram feature BG_GP_j=(bg₁,bg₂,…,bg_n)(n =1,2 ..., 32), the background area of particle is point centered by particle position, a width ofA height ofSquare Shape region deduct target rectangle region after " returning " font region, four outer dead centres E', F', G', H' of particle background area sit Mark is respectively as follows:

$(x_{E^{\&prime;}},y_{E^{\&prime;}})=(p\_x_j+\frac{\sqrt{2}}{2}height,p\_y_j+\frac{\sqrt{2}}{2}width),$

$(x_{F^{\&prime;}},y_{F^{\&prime;}})=(p\_x_j-\frac{\sqrt{2}}{2}height,p\_y_j+\frac{\sqrt{2}}{2}width),$

$(x_{G^{\&prime;}},y_{G^{\&prime;}})=(p\_x_j+\frac{\sqrt{2}}{2}height,p\_y_j-\frac{\sqrt{2}}{2}width),$

$(x_{H^{\&prime;}},y_{H^{\&prime;}})=(p\_x_j-\frac{\sqrt{2}}{2}height,p\_y_j-\frac{\sqrt{2}}{2}width),$

Then color histogram feature BG_HP of particle j background area_jWith local binary patterns histogram feature BG_GP_jIt is respectively as follows:

BG_HP_j=H_in(x_E',y_E')-H_in(x_G',y_G'-1)-H_in(x_F'-1,y_F')+H_in(x_H'-1,y_H'-1)-HP_j,

BG_GP_j=G_in(x_E',y_E')-G_in(x_G',y_G'-1)-G_in(x_F'-1,y_F')+G_in(x_H'-1,y_H'-1)-GP_j,

Calculate the likelihood ratio of the color characteristic of particle j and the color characteristic of its background areaWherein ε=0.001, if T₀The discrimination threshold value being characterized, Then the color characteristic of particle j with the discrimination of background color feature isThe face of particle j The feature deterministic coefficient of color characteristic isCalculate the local binary patterns feature of particle j and its The likelihood ratio of the local binary patterns feature of background area Wherein ε=0.001, the local binary patterns feature of particle j with the discrimination of background local binary patterns feature isThe feature deterministic coefficient of the local binary patterns feature of particle j is

A kind of block particle filter based on color and the anti-of local binary patterns Feature Fusion Method for tracking target, it is characterised in that: in described step 4, the subnormal color of dbjective state is melted with local binary patterns feature Close particle filter object tracking process be: p particle is calculated respectively particle j (j=1,2 ..., color characteristic HP p)_j= (hp₁,hp₂,…,hp_n) and color of object feature templates H=(h₁,h₂,…,h_n) (n=1,2 ..., 32) Pasteur's coefficientPasteur's distance isCalculate the color characteristic weight of particle jWherein σ=0.05, is normalized the color characteristic weight of each particleP particle is calculated respectively local binary patterns feature GP of particle j_j=(gp₁,gp₂,…,gp_n) With target local binary patterns feature templates G=(g₁,g₂,…,g_n) Pasteur's coefficientPasteur Distance isCalculate the local binary patterns feature weight of particle j Wherein σ=0.05, is normalized the local binary patterns feature weight of each particleFrom upper The feature deterministic coefficient of the color characteristic that can be calculated particle j in one step is β _ C_j, local binary patterns feature Feature deterministic coefficient is β _ L_j, calculating the weight after each particle color and local binary patterns Feature Fusion isIf (feature deterministic coefficient β _ C_j=β _ L_j=0, then make β _ C_j =β _ L_j=0.5), each particle weights is normalizedBy each particle coordinate by its Weight Obtain the center point coordinate of present frame (kth frame) target

A kind of block particle filter based on color and the anti-of local binary patterns Feature Fusion Method for tracking target, it is characterised in that: in described step 4, color when target is at least partially obscured and local binary patterns feature The particle filter object tracking process merged is: occur, according to target in previous frame image, the sub-block shape that circumstance of occlusion detects State flag bitP particle is calculated respectively particle j (j=1,2 ..., p) inTime Color histogram feature HP of the most effective sub-block_{j_i}With local binary patterns histogram feature GP_{j_i}；By each effective son of particle j The color characteristic of block and corresponding sub-block color characteristic template H_iContrast, calculate Pasteur's coefficient of each effective sub-block iTake the average of effective sub-block similarity as corresponding particle j integral part Similarity, remembers that effective sub-block number is M, then Pasteur's coefficient of particle j isPasteur's distance isCalculate the color characteristic weight of each particleColor to each particle Feature weight is normalizedAgain by the local binary patterns feature of each effective sub-block of particle j With corresponding sub-block local binary patterns feature templates G_iContrast, calculate Pasteur's coefficient of each effective sub-block iTake the average of effective sub-block similarity as corresponding particle j integral part Similarity, remembers that effective sub-block number is M, then Pasteur's coefficient of particle j isPasteur's distance isCalculate the local binary patterns feature weight of each particleTo each particle Local binary patterns feature weight is normalizedParticle can be calculated from previous step The feature deterministic coefficient of the color characteristic of j is β _ C_j, the feature deterministic coefficient of local binary patterns feature is β _ L_j, calculate each Weight after particle color and local binary patterns Feature Fusion is If (feature deterministic coefficient β _ C_j=β _ L_j=0, then make β _ C_j=β _ L_j=0.5), each particle weights is normalizedEach particle coordinate is obtained by its Weight the center point coordinate of present frame (kth frame) target

A kind of block particle filter based on color and the anti-of local binary patterns Feature Fusion Method for tracking target, it is characterised in that: the method for least square target prodiction in described step 4, when target is seriously blocked Process is: according to the target's center point coordinates (x of the most all frames_t,y_t) (t=1,2 ..., k-1), set up equation below:

$\left\{\begin{array}{c}{x}_t=a_{1}t+b_1\\ y_t=a_{2}t+b_2\end{array}\right.$

Each coefficient a is gone out by solving this Equation for Calculating₁,a₂,b₁,b₂, according to formula x_k=a₁k+b₁, y_k=a₂k+b₂It is calculated Center point coordinate (the x of target in k frame_k,y_k)。

A kind of block particle filter target based on color and the anti-of local binary patterns Feature Fusion Tracking, it is characterised in that: in described step 5, the renewal process of dbjective state is: can be calculated from preceding step and work as Center point coordinate (the x of front frame (kth frame) target_k,y_k), calculate color of object histogram feature H in present frame_acc=(h₁′, h₂′,……,h_n') (n=1,2 ..., 32), note present frame color of object feature and color characteristic template H=(h₁,h₂,…,h_n) (n=1,2 ..., 32) similarity beIf the overall similarity threshold value of target is T₁, when B is big In equal to threshold value T₁Time, illustrate that target is normal condition in the current frame, if now dbjective state flag bit Flag is equal to 0, Then keeping constant, otherwise updating current goal state flag bit is 0, i.e. shows that now target has had been detached from blocking；When B is less than Threshold value T₁Time, illustrate that target is blocked in the current frame, extract coordinates of targets (x_k,y_k) each sub-block i on region (i=1,2 ..., 6) color histogram feature is denoted as H_{acc_i}, calculate each sub-block color characteristic and corresponding sub-block color characteristic template H_iSimilarityIf the color characteristic similarity threshold of sub-block is T₂, then:

${\mathrm{Flag}}_{B^i}=\left\{\begin{array}{c}0,B^i<T_2\\ 1,B^i\&GreaterEqual;T_2\end{array}\right.$

I.e. work as BⁱLess than T₂, this sub-block i is invalid sub-block, remembers sub-block state flag bitIt is 0；Work as BⁱMore than or equal to T₂, should Sub-block i is effective sub-block, remembers sub-block state flag bitIt is 1, adds up the number M of effective sub-block, according to effective sub-block number Mesh judges the serious shielding degree of target:

$Flag=\left\{\begin{array}{c}1,M>2\\ 2,M\&le;2\end{array}\right.$

I.e. when the number M of effective sub-block is more than 2, illustrate that target is at least partially obscured in the current frame, update dbjective state mark Position Flag is 1, when the number M of effective sub-block is less than or equal to 2, illustrates that target is seriously blocked in the current frame, more fresh target State flag bit Flag is 2.

A kind of block particle filter based on color and the anti-of local binary patterns Feature Fusion Method for tracking target, it is characterised in that: in described step 6, template renewal method is: note color of object feature templates is H, currently The target new coordinates regional color histogram of frame is characterized as H_acc, then template renewal formula is: H=α H+ (1-α) H_acc, wherein, 0.80≤α≤0.99, the concrete numerical value of α sets according to video situation；The local binary patterns feature templates of target, the color of sub-block Feature templates, local binary patterns feature templates update method and the above-mentioned color of object feature templates update method class of sub-block With.