CN103413323B - Based on the object tracking methods of component-level apparent model - Google Patents

Abstract

The invention discloses a kind of object tracking methods based on component-level apparent model, it is the method upgrading apparent model based on middle level clue component-level, by carrying out super-pixel segmentation to image, the target component of tracking object is described with super-pixel, moving object in situations such as utilizing the information structuring object features pond of target component set up and upgrade the apparent model of object, this model can express deformation accurately, block.When upgrading object apparent model, the section components collection will be replaced in feature pool is replaced with the parts of new tracking frame, set up the feature supplementary set of target object parts, feature supplementary set is added feature pool as the component representation information of new tracking frame, then sets up according to new feature pool the renewal that apparent model realizes apparent model.This law, along with the carrying out to object tracking, upgrades the information that apparent model information concentrates tracked object, and object apparent model is more comprehensive, can obtain better effect when instructing the cosmetic variation such as tracking is seriously blocked, distortion larger.

Description

Based on the object tracking methods of component-level apparent model

Technical field

The present invention relates to computer vision field, more specifically, relate to a kind of object tracking methods based on component-level apparent model.

Background technology

Object tracking is an important research content of computer vision field, has caused the extensive concern of people in recent years, has become current study hotspot.This technology has broad application prospects, and plays an important role in multiple field, as security monitoring, man-machine interaction, medical diagnosis and vehicle flow monitoring etc.Although have already been proposed a large amount of object tracking methods, comparatively large or have when seriously blocking in illumination and object profile variation, these methods often can not provide desirable tracking results, and Chang Wufa tracks target object.Therefore, propose a kind of effective object tracking methods and there is important using value and realistic meaning.

Current, Bayesian filter principle is quite ripe on object tracking uses, and the feature extraction related to, sets up apparent model, search target, upgrades apparent model four major part, and wherein Focal point and difficult point is apparent model process.Although a lot of successfully object tracking algorithm is suggested, develop an energy process complexity, the robust algorithm of dynamic scene remains a challenging problem.Because illumination changes, camera lens is moved, object generation deformation, target object generating portion or all to block etc. the outward appearance of scene can be caused to change a lot.These changes can only be processed by the adaptive approach of their expression of incremental update.Therefore, it is possible to the online updating of unceasing study is necessary for the apparent expression of tracking object for tracing task.

Existing object tracking methods, when upgrading object apparent model, based on template, is upgrade unit with frame, namely in Modelling feature pond, adds a frame information and just from feature pool, reject a frame information.Such processing mode makes object apparent model constantly update along with the change of tracking target or scene outward appearance, but, part useful information can be caused to lose at rejecting one frame information.In target object deformation, frequent or part is when following the tracks of die-out time long enough in scene, and it is comprehensive that result upgrades the apparent model obtained often, can only obtain the part apparent to object and represent.Tracking under the background of complexity and the large scene of object cosmetic variation, do not have healthy and strong apparent model, Chang Wufa obtains effective tracking results.

Summary of the invention

In order to overcome the deficiencies in the prior art, namely for object tracking methods when based on template renewal object, apparent model easily causes by apparent model loss part tracking object information, model represents not comprehensive to tracking object, the present invention proposes a kind of object tracking methods based on component-level apparent model, be upgrade unit with parts, be intended to strengthen the integrality of apparent model to tracking object and represent.

In order to overcome the deficiencies in the prior art, technical scheme of the present invention is:

Based on an object tracking methods for component-level apparent model, comprise the following steps:

S1. create and be used for the feature pool of modeling: simple follow the tracks of before m two field picture record the target area of every frame, expand to surrounding the region that is expanded centered by target area, super-pixel splits each extended area, the information of super-pixel record object object part, extract the feature of each parts, and collect the feature construction feature pool of all frames;

S2. the feature set in feature based pond creates the apparent model of object;

S3. establish the tracking having completed front t two field picture, t >=m, calculate feature set and the degree of confidence of super-pixel in the target area of t+1 two field picture and extended area thereof according to apparent model, the super-pixel of record description target object parts;

S4. calculate the supplementary set of t+1 frame image features collection, when seriously not blocking, perform S5, otherwise perform S8;

S5. using feature pool middle distance current time frame at most as being replaced frame;

S6. when being replaced in frame the super-pixel quantity describing target object parts and being greater than β, β is the constant preset, from these super-pixel, select β super-pixel form a set, namely this set is the supplementary set of present frame feature set, wherein to describe the Euclidean distance of the eigenvector of the super-pixel of target maximum for the eigenvector of β super-pixel and present frame, then proceed to S11; Otherwise proceed to S7;

S7. using being replaced in frame the feature set of the super-pixel describing target object parts as the supplementary set of present frame feature set, S11 is proceeded to;

S8. the 3rd frame selecting feature pool middle distance current time nearest is as being replaced frame;

S9. when being replaced in frame the super-pixel quantity describing target object parts and being less than or equal to α, α is the constant preset, α < β, using the supplementary set of the feature set of these super-pixel as present frame feature set, then proceed to S11, otherwise, when being replaced in frame the super-pixel quantity describing target object parts and being greater than α, proceed to S10;

S10. using the supplementary set of the feature set of α maximum for the degree of confidence being replaced in frame the super-pixel describing a target object parts super-pixel as present frame feature set;

S11. merge the feature set of present frame and supplementary set thereof as the new feature collection of present frame, new feature collection added feature pool and deletes in feature pool the feature set being replaced frame, completing a feature pool and upgrade;

Upgrade apparent Model Condition if S12. meet, build the apparent model of object according to the feature pool after renewal, realize the renewal of apparent model;

S13. step S3 is proceeded to, until complete the tracking of whole sequence of video images.

Further, before in described step S1, m frame is the tracking based on instructing without apparent model, and the concrete mode creating feature pool is:

Given first two field picture Frame ₁middle target area, comprises central point and area size, with the target area of the first two field picture for template, with the alternative manner of simple match respectively from Frame ₂..., Frame _mmiddle calculating target area;

To sample alternatively target area in target area peripheral extent, to be doubly expanded region to surrounding expansion λ centered by target area, and respectively super-pixel is carried out to the extended area of m frame and be divided into N _iindividual super-pixel sp (i, j), i=1 ..., m, j=1 ..., N _i;

Extract the HSI color character of each frame super-pixel respectively, use eigenvector f _i ^jrepresent, and record each super-pixel and whether belong to target component; Finally, m frame feature vector is organized into the feature pool for creating parts apparent model in order $F=\left\{f_{t^{\&prime;}}^r\right|t^{\&prime;}=1,...,m;r=1,...,N_{t^{\&prime;}}\},$ And record.

Above-mentioned λ is constant, in order to ensure that extended area is enough large, should cover each sampling; Because HSI color space is closer to human eye vision state, consistent with the understanding of human eye vision to target component, so extract the HSI color character of each frame super-pixel.

Further, the apparent model creating object in described step S2 comprises by means Method the feature vector cluster in feature pool and each cluster degree of confidence two parts of calculating, adopt a base part of each cluster representative feature similarity, and represent that parts are the probability of target component with confidence value; Be implemented as follows:

According to means clustering algorithm to the eigenvector in feature pool be clustered into n class clst (k) (k=1 ..., n), f _crepresent cluster centre eigenvector, r _ck () is the radius of cluster clst (k) in feature space;

If S ⁺k () covers the area summation in target area for the parts belonging to a kth cluster in feature pool, S ^-k () covers the area summation outside target area for the parts belonging to a kth cluster in feature pool, then the degree of confidence of cluster is expressed as: $C_k=\frac{S^{+}\left(k\right)-S^{-}\left(k\right)}{S^{+}\left(k\right)+S^{-}\left(k\right)},\&ForAll;k=1,...,n.$

Further, described step S3 establishes the tracking completing front t two field picture, t >=m, and calculate feature set and the degree of confidence of super-pixel in the target area of t+1 two field picture and extended area thereof according to apparent model, concrete grammar is as follows:

Carry out super-pixel segmentation to the central point of the target area of t frame and size at the extended area of t+1 two field picture, and extract the HSI feature of each super-pixel, use eigenvector represent;

Eigenvector carries out similarity system design with the eigenvector in feature pool respectively, determines the corresponding relation between the super-pixel of t+1 frame and cluster by the corresponding relation of eigenvector in feature pool and cluster;

If λ _dconstant, super-pixel sp (t+1, j) belongs to cluster clst (k), then with cluster centre eigenvector f _ck the cluster weight of () is: $dist(j,k)=\exp (-{\mathrm{\&lambda;}}_d\&times;\frac{\left|\right|f_{t+1}^j-f_c\left(k\right)\left|\right|}{r_c\left(k\right)}),$ The degree of confidence of super-pixel sp (t+1, j) is conf (t+1, j)=dist (j, k) × C _k, and record the degree of confidence of super-pixel; Draw the degree of confidence figure of extended area, the value of each point on figure is corresponding super-pixel confidence value;

M is adopted in extended area _t+1individual sample as t+1 frame candidate target region, M can be obtained by the corresponding relation of extended area and degree of confidence figure _t+1the degree of confidence of individual sample, and according to maximum a-posteriori estimation using degree of confidence in each candidate target region and maximum as target area, and the super-pixel of record description target object parts.

Further, described step S4 to step S11 adopts in keeping characteristics pond the Partial Feature that is replaced frame to add present frame as the supplementary set of the feature set of present frame, then replaces the regeneration characteristics pond method being replaced frame by new present frame feature set; The criterion of seriously blocking in described step S4 is: establish θ _ofor occlusion threshold, when candidate target degree of confidence is less than θ _oduring product with extended area, be judged as there occurs and seriously block.

Easily lose for replacing unit the super-pixel feature that part describes target component with whole frame for when regeneration characteristics pond in step S4 to step S11, adopt in keeping characteristics pond the Partial Feature that is replaced frame to add present frame as the supplementary set of the feature set of present frame, then replace the regeneration characteristics pond method being replaced frame by new present frame feature set.Seriously not blocking two kinds of situations for seriously blocking and, adopting different strategies when calculating supplementary set.The tactful advantage in this regeneration characteristics pond has: the first, strengthens the degree of confidence with the super-pixel being retained the same cluster of super-pixel; Second, retain to cause because target appearance changes same parts in the feature being replaced in frame and new frame the super-pixel having different description, or make not have at new frame because blocking and be replaced in frame the feature having the super-pixel being described as target component, enrich the description to parts in feature pool, make object apparent model more comprehensive.

Further, the apparent model after upgrading in described step S12 comprises cluster and calculates degree of confidence, creates apparent model and adopts the modeling of m frame: according to mean shift clustering algorithm to the eigenvector in feature pool carry out cluster, be clustered into n class clst (k) (k=1 ..., n), each cluster centre is f _ck (), the member of each class is expressed as calculating cluster confidence value part, calculating by preserving the super-pixel information that each two field picture is split and the super-pixel information of supplementing calculated as each frame in feature pool: establish be any one feature in feature supplementary set, it belongs to a kth cluster, and the area of super-pixel corresponding to it is Area (t', o), S ⁺k () covers the area summation in target area for the super-pixel belonging to a kth cluster in feature pool: S ⁺(k)=∑ N ⁺(t', r)+∑ Area (t', o),

s ^-k () covers the area summation outside target area for the super-pixel belonging to a kth cluster in feature pool: S ^-(k)=∑ N ^-(t', r), cluster confidence value is: $C_k=\frac{S^{+}\left(k\right)-S^{-}\left(k\right)}{S^{+}\left(k\right)+S^{-}\left(k\right)},\&ForAll;k=1,...,n.$

Compared with prior art, the present invention has following beneficial effect:

1) middle level clue, can represent image result information more effectively, neatly.Target part that is significant for tool in target, that have obvious boundary information is slit into numerous super-pixel, and then describes the parts of tracking object by super-pixel, operate more directly perceived.

2) take parts as minimum operation unit, replace at the most close parts of selection each time, retain the supplementary set that the most dissimilar parts are a new frame parts collection, the information making the object in modeling information pond apparent is abundanter, and apparent model can to the description of tracking object more comprehensively.

Accompanying drawing explanation

Fig. 1 is that method of the present invention performs step schematic diagram.

Fig. 2 is the inventive method and is the method effect contrast figure when tracking image sequence " Wwoman_sequence " the 63rd frame figure upgrading unit with frame.

Fig. 3 is the inventive method and is the method effect contrast figure when tracking image sequence " Wwoman_sequence " the 78th frame figure upgrading unit with frame.

Embodiment

Below in conjunction with accompanying drawing, the present invention will be further described, but embodiments of the present invention are not limited to this.

Method of the present invention performs step schematic diagram as shown in Figure 1, specifically comprises the steps:

S1. create the feature pool stage being used for modeling: m frame before simple tracking, in the present embodiment, m gets 7; Image also records the target area of every frame, first, and given first two field picture Frame ₁middle target area, comprises central point and area size, with the target area of the first two field picture for template, with the alternative manner of simple match respectively from Frame ₂..., Frame ₇middle calculating target area; Then, centered by target area to surrounding expansion λ doubly, λ is constant, and in the present embodiment, λ gets 1.5, in order to ensure that extended area is enough large, should cover each sampling; Be expanded region, and use SLIC algorithm to carry out super-pixel to the extended area of 7 frames to be respectively divided into N _iindividual super-pixel sp (i, j) (i=1 ..., 7, j=1 ..., N _i); Then, extract the HSI color character of each frame super-pixel respectively, use eigenvector f _i ^jrepresent, and establish N ⁺represent super-pixel pixel number in target area, N ^-represent that super-pixel is at target area exterior pixel number, passes through N ⁺/ (N ^-+ N ⁺) value judge super-pixel affiliated area, is on dutyly greater than 0.5 and records this super-pixel and belong in target area, otherwise record super-pixel belongs to outside region; Finally, 7 frame feature vectors are organized into the feature pool for creating parts apparent model in order $F=\left\{f_{t^{\&prime;}}^r\right|t^{\&prime;}=1,...,7;r=1,...,N_{t^{\&prime;}}\},$ And record.

S2. the initial apparent model order section of target is created: first, according to means clustering algorithm to the eigenvector in feature pool be clustered into n class clst (k) (k=1 ..., n), use f _crepresent cluster centre eigenvector, r _ck () is the radius of cluster clst (k) in feature space.Then, if S ⁺k () covers the area summation in target area for the parts belonging to a kth cluster in feature pool, S ^-k () covers the area summation outside target area for the parts belonging to a kth cluster in feature pool, then the degree of confidence of cluster is expressed as: $C_k=\frac{S^{+}\left(k\right)-S^{-}\left(k\right)}{S^{+}\left(k\right)+S^{-}\left(k\right)},\&ForAll;k=1,...,n.$

S3. based on the image trace target object stage of apparent model to new input: first, carry out super-pixel segmentation to the central point of the target area of t frame and size at the extended area of t+1 two field picture, and extract the HSI feature of each super-pixel, use eigenvector represent; Then, eigenvector carries out similarity system design with the eigenvector in feature pool respectively, determines the corresponding relation between the super-pixel of t+1 frame and cluster by the corresponding relation of eigenvector in feature pool and cluster; Thus, if λ _dconstant, in the present embodiment, λ _dget 2, super-pixel sp (t+1, j) belongs to cluster clst (k), then with cluster centre eigenvector f _ck the cluster weight of () is: $dist(j,k)=\exp (-{\mathrm{\&lambda;}}_d\&times;\frac{\left|\right|f_{t+1}^j-f_c\left(k\right)\left|\right|}{r_c\left(k\right)}),$ And then the degree of confidence of super-pixel sp (t+1, j) is conf (t+1, j)=dist (j, k) × C _k, and record the confidence value of super-pixel; Then, draw the degree of confidence figure of extended area, the value of each point on figure is corresponding super-pixel confidence value; Finally, in extended area, M is adopted _t+1individual sample as t+1 frame candidate target region, M can be obtained by the corresponding relation of extended area and degree of confidence figure _t+1the degree of confidence of individual sample, and according to maximum a-posteriori estimation using degree of confidence in each candidate target region and maximum as target area, and record super-pixel and whether belong to object judgement.

S4. present frame feature is replaced in feature pool and is replaced frame feature stage: first will define in this stage and serious shadowing standard occurs: establish θ _ofor occlusion threshold, in the present embodiment, for Wwoman_sequence image sequence, θ _oget-0.1; When candidate target degree of confidence is less than θ _oduring product with extended area, be judged as there occurs and seriously block.When occurring seriously to block, forward step S6 to, otherwise, forward step S5 to.

S5. under non-serious circumstance of occlusion, present frame feature local is replaced in feature pool and is replaced frame feature stage: using feature pool middle distance current time frame at most as being replaced frame.If β, β is the constant preset, in the present embodiment, β gets 25, when being replaced in frame the super-pixel quantity describing target object parts and being greater than β, from these super-pixel, select β super-pixel form a set, namely this set is the supplementary set of present frame feature set, and wherein to describe the Euclidean distance of the eigenvector of the super-pixel of target maximum for the eigenvector of this β super-pixel and present frame; Otherwise, will the supplementary set of feature set as present frame feature set of the super-pixel describing target object parts be replaced in frame.Forward step S7 to.

S6. under serious circumstance of occlusion, present frame feature local is replaced in feature pool and is replaced frame feature stage: select nearest 3rd frame of feature pool middle distance current time as being replaced frame.If α, α are the constant preset, wherein α < β, in the present embodiment, α gets 15, when being replaced in frame the super-pixel quantity describing target object parts and being less than α, using the supplementary set of the feature set of these super-pixel as present frame feature set; Otherwise, using the supplementary set of the feature set of α maximum for the value of the confidence being replaced in frame the super-pixel describing a target object parts super-pixel as present frame feature set.

S7. in the regeneration characteristics pond stage: first, merge the feature set of present frame and supplementary set thereof the new feature collection as present frame, and the super-pixel feature in supplementary set is recorded as the super-pixel feature of the description target object parts of present frame; Then, new feature collection is added feature pool, and delete in feature pool the feature set being replaced frame.

S8. judge whether to meet and upgrade apparent Model Condition, when not meeting, forward step S3 to, otherwise, continue step S9;

S10. apparent model order section is upgraded: first, according to mean shift clustering algorithm to the eigenvector in feature pool $(F=\{f_{t^{\&prime;}}^r|t^{\&prime;}=1,...,m;r=1,...,N_{t^{\&prime;}}\})$ Carry out cluster, be clustered into n class clst (k) (k=1 ..., n), each cluster centre is f _ck (), the member of each class is expressed as then, cluster confidence value is calculated.Because the super-pixel information of each two field picture segmentation of preserving in feature pool adds supplementary set, therefore, the method calculating cluster the value of the confidence needs corresponding change: establish be any one feature in feature supplementary set, it belongs to a kth cluster, and the area of super-pixel corresponding to it is Area (t', o), S ⁺k () covers the area summation in target area for the super-pixel belonging to a kth cluster in feature pool: S ⁺(k)=∑ N ⁺(t', r)+∑ Area (t', o), s ^-k () covers the area summation outside target area for the super-pixel belonging to a kth cluster in feature pool: S ^-(k)=∑ N ^-(t', r), cluster confidence value is: $C_k=\frac{S^{+}\left(k\right)-S^{-}\left(k\right)}{S^{+}\left(k\right)+S^{-}\left(k\right)},\&ForAll;k=1,...,n.$

If S11. complete the tracking of whole sequence of video images, terminate program, otherwise, forward step S3 to.

Fig. 2 is adopt the inventive method to be upgrade unit with parts and is the method effect contrast figure when tracking image sequence " Wwoman_sequence " the 63rd frame figure upgrading unit with frame.Fig. 2 (a), Fig. 2 (b), Fig. 2 (c) are the implementation of the method taking frame as renewal unit; The implementation that Fig. 2 (d), Fig. 2 (e), Fig. 2 (f) are this method.Fig. 2 (a) and Fig. 2 (d) carries out super-pixel segmentation figure to the extended area of target area; Fig. 2 (b) and Fig. 2 (e) assesses super-pixel in extended area for using apparent model, draw out degree of confidence figure, employing gray-scale map represents, wherein the degree of confidence of dark parts (black) is greater than 0, is identified as tracking target; On the contrary, the confidence value of light color (grey) part is less than 0, is identified as background; Fig. 2 (c) and Fig. 2 (f) is the result of following the tracks of, and frame is target frame.In Fig. 2, tracking target entering part blocks the tracking under environment, can in two ways can well modeling to the viewable portion of target, apparent model is to the viewable portion of tracking target all identifiable design, and in degree of confidence figure, the confidence value of the super-pixel of tracking target is greater than 0.Fig. 3 is adopt the inventive method to be upgrade unit with parts and is the method effect contrast figure when tracking image sequence " Wwoman_sequence " the 78th frame figure upgrading unit with frame.Fig. 3 (a), Fig. 3 (b), Fig. 3 (c) are the implementation of the method taking frame as renewal unit; The implementation that Fig. 3 (d), Fig. 3 (e), Fig. 3 (f) are this method.Fig. 3 (a) and Fig. 3 (d) carries out super-pixel segmentation figure to the extended area of target area; Fig. 3 (b) and Fig. 3 (e) assesses super-pixel in extended area for using apparent model, draw out degree of confidence figure, employing gray-scale map represents, wherein the degree of confidence of dark parts (black) is greater than 0, is identified as tracking target; On the contrary, the confidence value of light color (grey) part is less than 0, is identified as background; Fig. 3 (c) and Fig. 3 (f) is tracking results, and frame is target frame.In Fig. 3, the tracking target part that is blocked starts to walk out the tracking of blocking, and through the tracking of multiframe, the update method in units of frame has been blocked tracked target the characteristic loss of part, to the part not identifiable design that is blocked; On the contrary, the inventive method carries out the information of local updating replacement for the information pool of modeling in units of parts, the information of more complete reservation tracking target, apparent model can than more comprehensive expression to tracking target, therefore, although tracking target lower part is blocked more than m frame, can be identified when this part is walked out and blocked environment, as the leg portion of Fig. 3 (b) personage, dotted box portion in figure.

Above-described embodiments of the present invention, do not form limiting the scope of the present invention.Any amendment done within spiritual principles of the present invention, equivalent replacement and improvement etc., all should be included within claims of the present invention.

Claims (2)

1. based on an object tracking methods for component-level apparent model, it is characterized in that, comprise the following steps:

S1. create and be used for the feature pool of modeling: m two field picture before following the tracks of also records the target area of every frame, expand to surrounding the region that is expanded centered by target area, super-pixel splits each extended area, the information of super-pixel record object object part, extract the feature of each parts, and collect the feature construction feature pool of all frames;

S2. the feature set in feature based pond creates the apparent model of object;

S3. establish the tracking having completed front t two field picture, t >=m, calculate feature set and the degree of confidence of super-pixel in the target area of t+1 two field picture and extended area thereof according to apparent model, the super-pixel of record description target object parts;

S4. calculate the supplementary set of t+1 frame image features collection, when seriously not blocking, perform S5, otherwise perform S8;

S5. using feature pool middle distance current time frame at most as being replaced frame;

S6. when being replaced in frame the super-pixel quantity describing target object parts and being greater than β, β is the constant preset, from these super-pixel, select β super-pixel form a set, namely this set is the supplementary set of present frame feature set, as the supplementary set of present frame feature set, wherein to describe the Euclidean distance of the eigenvector of the super-pixel of target maximum for the eigenvector of β super-pixel and present frame, then proceed to S11; Otherwise proceed to S7;

S7. using being replaced in frame the feature set of the super-pixel describing target object parts as the supplementary set of present frame feature set, S11 is proceeded to;

S8. the 3rd frame selecting feature pool middle distance current time nearest is as being replaced frame;

S9. when being replaced in frame the super-pixel quantity describing target object parts and being less than or equal to α, α is the constant preset, α < β, using the supplementary set of the feature set of these super-pixel as present frame feature set, then proceed to S11, otherwise, when being replaced in frame the super-pixel quantity describing target object parts and being greater than α, proceed to S10;

S10. using the supplementary set of the feature set of α maximum for the degree of confidence being replaced in frame the super-pixel describing a target object parts super-pixel as present frame feature set;

S11. merge the feature set of present frame and supplementary set thereof as the new feature collection of present frame, new feature collection added feature pool and deletes in feature pool the feature set being replaced frame, completing a feature pool and upgrade;

Upgrade apparent Model Condition if S12. meet, build the apparent model of object according to the feature pool after renewal, realize the renewal of apparent model;

S13. step S3 is proceeded to, until complete the tracking of whole sequence of video images;

Before in described step S1, m frame is the tracking based on instructing without apparent model, and the concrete mode creating feature pool is:

Given first two field picture Frame ₁middle target area, comprises central point and area size, with the target area of the first two field picture for template, with the alternative manner of simple match respectively from Frame ₂..., Frame _mmiddle calculating target area;

To sample alternatively target area in target area peripheral extent, to be doubly expanded region to surrounding expansion λ centered by target area, and respectively super-pixel is carried out to the extended area of m frame and be divided into N _iindividual super-pixel sp (i, j), i=1 ..., m, j=1 ..., N _i;

Extract the HSI color character of each frame super-pixel respectively, use eigenvector f _i ^jrepresent, and record each super-pixel and whether belong to target component; Finally, m frame feature vector is organized into the feature pool for creating parts apparent model in order $F=\left\{f_{t^{\&prime;}}^r\right|t^{\&prime;}=1,...,m;r=1,...,N_{t^{\&prime;}}\},$ And record;

The apparent model creating object in described step S2 comprises by means Method the feature vector cluster in feature pool and each cluster degree of confidence two parts of calculating, adopt a base part of each cluster representative feature similarity, and represent that parts are the probability of target component with confidence value; Be implemented as follows:

According to means clustering algorithm to the eigenvector in feature pool $F=\left\{f_{t^{\&prime;}}^r\right|t^{\&prime;}=1,...,m;r=1,...,N_{t^{\&prime;}}\}$ Be clustered into n class clst (k), k=1 ..., n, f _crepresent cluster centre eigenvector, r _ck () is the radius of cluster clst (k) in feature space;

If S ⁺k () covers the area summation in target area for the parts belonging to a kth cluster in feature pool, S ^-k () covers the area summation outside target area for the parts belonging to a kth cluster in feature pool, then the degree of confidence of cluster is expressed as: $C_k=\frac{S^{+}\left(k\right)-S^{-}\left(k\right)}{S^{+}\left(k\right)+S^{-}\left(k\right)},\&ForAll;k=1,...,n;$

Described step S3 establishes the tracking completing front t two field picture, t >=m, and calculate feature set and the degree of confidence of super-pixel in the target area of t+1 two field picture and extended area thereof according to apparent model, concrete grammar is as follows:

Carry out super-pixel segmentation to the central point of the target area of t frame and size at the extended area of t+1 two field picture, and extract the HSI feature of each super-pixel, use eigenvector represent;

Eigenvector carries out similarity system design with the eigenvector in feature pool respectively, determines the corresponding relation between the super-pixel of t+1 frame and cluster by the corresponding relation of eigenvector in feature pool and cluster;

If λ _dconstant, super-pixel sp (t+1, j) belongs to cluster clst (k), then with cluster centre eigenvector f _ck the cluster weight of () is: $dist(j,k)=\exp (-{\mathrm{\&lambda;}}_d\&times;\frac{\left|\right|f_{t+1}^j-f_c\left(k\right)\left|\right|}{r_c\left(k\right)}),$ The degree of confidence of super-pixel sp (t+1, j) is conf (t+1, j)=dist (j, k) × C _k, and record the degree of confidence of super-pixel; Draw the degree of confidence figure of extended area, the value of each point on figure is corresponding super-pixel confidence value;

M is adopted in extended area _t+1individual sample as t+1 frame candidate target region, M can be obtained by the corresponding relation of extended area and degree of confidence figure _t+1the degree of confidence of individual sample, and according to maximum a-posteriori estimation using degree of confidence in each candidate target region and maximum as target area, and the super-pixel of record description target object parts;

Described step S4 to step S11 adopts in keeping characteristics pond the Partial Feature that is replaced frame to add present frame as the supplementary set of the feature set of present frame, then replaces the regeneration characteristics pond method being replaced frame by new present frame feature set; The criterion of seriously blocking in described step S4 is: establish θ _ofor occlusion threshold, when candidate target degree of confidence is less than θ _oduring product with extended area, be judged as there occurs and seriously block.

2. the object tracking methods based on component-level apparent model according to claim 1, it is characterized in that, apparent model after upgrading in described step S12 comprises cluster and calculates degree of confidence, adopts the modeling of m frame: according to mean shift clustering algorithm to the eigenvector in feature pool carry out cluster, be clustered into n class clst (k), k=1 ..., n, each cluster centre is f _ck (), the member of each class is expressed as calculating cluster confidence value part, calculating by preserving the super-pixel information that each two field picture is split and the super-pixel information of supplementing calculated as each frame in feature pool: establish be any one feature in feature supplementary set, it belongs to a kth cluster, and the area of super-pixel corresponding to it is Area (t', o), S ⁺k () covers the area summation in target area for the super-pixel belonging to a kth cluster in feature pool:

S ⁺(k)＝∑N ⁺(t',r)+∑Area(t',o)，

s ^-k () covers the area summation outside target area for the super-pixel belonging to a kth cluster in feature pool: S ^-(k)=∑ N ^-(t', r), cluster confidence value is: $C_k=\frac{S^{+}\left(k\right)-S^{-}\left(k\right)}{S^{+}\left(k\right)+S^{-}\left(k\right)},\&ForAll;k=1,...,n,N^{+}$ Represent super-pixel pixel number in target area, N ^-represent that super-pixel is at target area exterior pixel number, o represents o super-pixel.