CN101996312B - Method and device for tracking targets - Google Patents

Abstract

The invention relates to a method and device for tracking targets. The method comprises the following steps: calculating a plurality of similarity curved surfaces related to a plurality of characteristics of the targets, wherein the similarity curved surfaces are curved surfaces which are formed by similarity between a group of candidate targets and target templates; calculating reliability of the characteristics based on parameters related to the similarity curved surfaces and parameters referring to the similarity curved surfaces, wherein the referred similarity curved surfaces are corresponding similarity curved surfaces obtained when an object is in an initial position, the related similarity curved surface parameters and the parameters referring to the similarity curved surfaces reflect the reliability of the corresponding characteristic, the parameters include one or more of the flat degree of the similarity curved surfaces, average similarity and multimodal degree; weighing and adding the plurality of similarity curved surfaces based on the reliability of each characteristic to obtain the synthesized similarity curved surfaces, wherein the characteristic with large reliability is given more weight; and determining the position of the targets based on the synthesized similarity curved surfaces. The method and device for tracking the targets of the invention can effectively track the targets.

Description

The method and apparatus of tracking target

Technical field

The application relates to computer vision field, relates more specifically to the tracking of target.

Background technology

Target following is a major issue of computer vision field, and it is behavioural analysis, and the basis of the various higher layer applications such as abnormality detection, has great significance in video monitoring system.

The factor affecting target following effect is a lot, the change of wherein illumination, the change of dbjective state, and to block (comprise between target block and target is blocked by background) be most important three kinds of situations.In order to accurate robustly tracking target in video, people introduce various feature, such as color histogram, gradient orientation histogram (HOG), Haar wavelet character, edge feature, LBP (Local BinaryPattern, local binary patterns) feature etc.Owing to respectively having relative merits between feature, the problem that Fusion Features has just become target tracking domain one important.

Summary of the invention

Provide hereinafter about brief overview of the present invention, to provide about the basic comprehension in some of the present invention.Should be appreciated that this general introduction is not summarize about exhaustive of the present invention.It is not that intention determines key of the present invention or pith, and nor is it intended to limit the scope of the present invention.Its object is only provide some concept in simplified form, in this, as the preorder in greater detail discussed after a while.

One object of the present invention is the method and apparatus providing a kind of new tracking target.

According to an aspect of the present invention, a kind of method of tracking target comprises: calculate multiple similarity curved surfaces relevant to multiple features of target respectively, and this similarity curved surface is the curved surface that one group of similarity between candidate target and To Template is formed; The confidence level of feature is calculated based on the parameter of relevant similarity curved surface and the parameter of reference similarity curved surface, wherein with reference to similarity curved surface be target be in initial position time the corresponding similarity curved surface that obtains, the confidence level of the parameter of relevant similarity curved surface and the parameter reflection individual features with reference to similarity curved surface; Multiple similarity curved surface is weighted addition to obtain synthesis similarity curved surface by the confidence level based on each feature, and the feature that wherein confidence level is large is endowed larger weight; and the position of target is determined based on synthesis similarity curved surface, the confidence level wherein calculating feature comprises the planarization based on relevant similarity curved surface, one or more in average similarity and multimodal degree and the planarization with reference to similarity curved surface, one or more confidence levels calculating feature in average similarity and multimodal degree, the number that the planarization of wherein relevant similarity curved surface and reference similarity curved surface is greater than the point of predetermined threshold based on similarity in corresponding similarity curved surface is respectively determined, relevant similarity curved surface and the average similarity with reference to similarity curved surface are the average similarity that in corresponding similarity curved surface, similarity is greater than the point of predetermined threshold respectively, relevant similarity curved surface and be the value on the second peak in corresponding similarity curved surface respectively and the ratio of the value at top with reference to the multimodal degree of similarity curved surface, the step wherein calculating the confidence level of feature comprises one or more in the following manner to determine confidence level: confidence level and planarization negative correlation, confidence level and average similarity positive correlation, confidence level and multimodal degree negative correlation, the step wherein calculating the confidence level of feature comprises the confidence level being calculated feature by following formula:

$\mathrm{Conf}=\frac{e^{-\mathrm{c\γ}}{e}^{-k(1-\min (1,\stackrel{\‾}{d}/\stackrel{\‾}{d_0}))}}{S/S_0}$

The wherein confidence level of Conf representation feature, represent the average similarity of relevant similarity curved surface, represent the average similarity with reference to similarity curved surface, S represents the planarization of relevant similarity curved surface, S ₀represent the planarization with reference to similarity curved surface, γ represents the multimodal degree of relevant similarity curved surface, c, k be more than or equal to 0 constant, represent 1 He in less one.

According to a further aspect in the invention, a kind of device of tracking target comprises: similarity curved surface computing unit, be configured to calculate multiple similarity curved surfaces relevant to multiple features of target respectively, this similarity curved surface is the curved surface that one group of similarity between candidate target and To Template is formed; Confidence level computing unit, be configured to the confidence level calculating feature based on the parameter of relevant similarity curved surface and the parameter of reference similarity curved surface, wherein with reference to similarity curved surface be target be in initial position time the corresponding similarity curved surface that obtains, the confidence level of the parameter of relevant similarity curved surface and the parameter reflection individual features with reference to similarity curved surface; Similarity curved surface synthesis unit, multiple similarity curved surface is weighted addition to obtain synthesis similarity curved surface by the confidence level be configured to based on each feature, and the feature that wherein confidence level is large is endowed larger weight; and target location determining unit, be configured to the position determining target based on synthesis similarity curved surface, wherein confidence level computing unit is configured to the confidence level calculating feature based on the parameter of relevant similarity curved surface and the parameter of reference similarity curved surface, wherein confidence level computing unit is configured to the planarization based on relevant similarity curved surface, one or more and the described planarization with reference to similarity curved surface in average similarity and multimodal degree, one or more confidence levels calculating feature in average similarity and multimodal degree, the number that the planarization of wherein relevant similarity curved surface and reference similarity curved surface is greater than the point of predetermined threshold based on similarity in corresponding similarity curved surface is respectively determined, relevant similarity curved surface and the average similarity with reference to similarity curved surface are the average similarity that in corresponding similarity curved surface, similarity is greater than the point of predetermined threshold respectively, relevant similarity curved surface and be the value on the second peak in corresponding similarity curved surface respectively and the ratio of the value at top with reference to the multimodal degree of similarity curved surface, wherein confidence level computing unit is configured to one or more in the following manner to determine confidence level: confidence level and planarization negative correlation, confidence level and average similarity positive correlation, confidence level and multimodal degree negative correlation, wherein confidence level computing unit is configured to the confidence level being calculated feature by following formula:

$\mathrm{Conf}=\frac{e^{-\mathrm{c\γ}}{e}^{-k(1-\min (1,\stackrel{\‾}{d}/\stackrel{\‾}{d_0}))}}{S/S_0}$

The wherein confidence level of Conf representation feature, represent the average similarity of relevant similarity curved surface, represent the average similarity with reference to similarity curved surface, S represents the planarization of relevant similarity curved surface, S ₀represent the planarization with reference to similarity curved surface, γ represents the multimodal degree of relevant similarity curved surface, c, k be more than or equal to 0 constant, represent 1 He in less one.

According to a further aspect of the invention, additionally provide a kind of computer program, for realizing following methods: calculate multiple similarity curved surfaces relevant to multiple features of target respectively, this similarity curved surface is the curved surface that one group of similarity between candidate target and To Template is formed; Parameter based on relevant similarity curved surface calculates the confidence level of feature; Multiple similarity curved surface carries out synthesizing to obtain synthesis similarity curved surface by the confidence level based on each feature; And the position of target is determined based on synthesis similarity curved surface.

According to a further aspect of the invention, additionally providing a kind of computer program of at least computer-readable medium form, it recording the computer program code of the method for realizing above-mentioned tracking target.

The present invention can tracking target effectively.

Accompanying drawing explanation

With reference to below in conjunction with the explanation of accompanying drawing to embodiment of the present invention, above and other objects, features and advantages of the present invention can be understood more easily.Parts in accompanying drawing are just in order to illustrate principle of the present invention.In the accompanying drawings, same or similar technical characteristic or parts will adopt same or similar Reference numeral to represent.

Fig. 1 shows the schematic diagram of Fusion Features according to the present embodiment;

Fig. 2 shows the process flow diagram of the method for the tracking target according to the first embodiment;

Fig. 3 shows the schematic diagram of the method for tracking target second embodiment of the invention;

The schematic diagram of the multi-peaks structure that the similarity curved surface that Fig. 4 shows HOG feature presents;

Fig. 5 shows the schematic diagram of the method for the tracking target according to the 3rd embodiment of the present invention;

Fig. 6 shows the schematic diagram of the method for the tracking target according to the 4th embodiment of the present invention;

Fig. 7 shows the schematic diagram of the method for the tracking target according to the 5th embodiment of the present invention;

Fig. 8 illustrates the structure of the citing of the computing equipment of the device of the tracking target that may be used for realizing embodiments of the present invention;

Fig. 9 shows the device of the tracking target according to the 6th embodiment of the present invention.

Embodiment

With reference to the accompanying drawings embodiments of the present invention are described.The element described in an accompanying drawing of the present invention or a kind of embodiment and feature can combine with the element shown in one or more other accompanying drawing or embodiment and feature.It should be noted that for purposes of clarity, accompanying drawing and eliminate expression and the description of unrelated to the invention, parts known to persons of ordinary skill in the art and process in illustrating.

the method of tracking target

first embodiment

Fig. 1 shows the schematic diagram of Fusion Features according to the present embodiment.Shown in Fig. 1 is fusion to two-way feature.The similarity curved surface of fisrt feature and the similarity curved surface of second feature are calculated respectively for the target (being pedestrian shown in Fig. 1) in video, then the similarity curved surface of two-way is weighted summation, is finally synthesized similarity curved surface.Thus obtain the maximum position of target probability of occurrence.To be further explained above-mentioned principle below.

In order to utilize the complementarity of different features, at least one global characteristics and at least one local feature in multiple feature, can be comprised.Global characteristics can be color histogram (Histogram of Color, HC), and local feature can be HOG, Haar wavelet character, edge feature, local binary patterns feature (Local Binary Pattern, LBP).Such as, the major advantage of HOG feature is insensitive to illumination variation, but its not discriminate individuals, separating capacity is not had for different pedestrian.And color histogram feature generally can discriminate individuals, is but often easily subject to the impact of illumination variation.

Fig. 2 shows the process flow diagram of the method for the tracking target according to the first embodiment.

Assuming that known target template and initial position thereof.This To Template and initial position thereof can be manual marks or be obtained by a reliable object detector.Such as, can use through training HOG (the Histogram of Oriented Gradients obtained, gradient orientation histogram) characteristic sum SVM (Support Vector Machine) sorter (hereinafter referred to as HOG sorter) to detect in video the initial target occurred as object detector, be defined as To Template with the target this detected, position target initially occurred is defined as initial position.Because target generally occupies certain area in the picture, conveniently, also the center (also can be other specified points, such as peak etc.) of target is called the position of target below.With the example that is fused to of two-way feature, present embodiment is introduced below, it should be understood that present embodiment also can be applied to situation about merging three or more features.

In step 202., similarity curved surface relevant with second feature to the fisrt feature of target is respectively calculated.

Similarity curved surface refers to the curved surface that one group of similarity between candidate target and To Template is formed.

Candidate target can be determined by the following method.In the process of following the tracks of, for present frame, based on the position of the target determined in former frame (when being initial frame for former frame, as mentioned above, the position of target by manual mark or can be obtained by a reliable object detector) and the target maximum translational speed estimated, judge the region that target may occur.Below the region that the target of judgement may occur is called region of search.Region of search determines the coordinate range of similarity curved surface.Region of search can be the target location determined with former frame is the center of circle, be multiplied by the circle that distance that the mistiming between two frames determines is radius with the target maximum translational speed estimated.It is square that region of search can also be that centered by the target location that former frame is determined, with the target translational speed estimated, be multiplied by distance that the mistiming between two frames determines be the length of side.The target maximum translational speed of above-mentioned estimation can be empirical value, or by the determined target of front cross frame position between distance be multiplied by a coefficient again divided by the value that the mistiming between two frames obtains.More simply, region of search can also be the target location determined with former frame for the center of circle, take preset distance as the circle of radius, or centered by the target location that former frame is determined, take preset distance as the square of the length of side.Those skilled in the art also it is contemplated that the method for other determination region of search.The size and shape of candidate target can be identical with To Template.Or the size of candidate target can carry out convergent-divergent along with the change of the distance between the target estimated and camera lens relative to To Template.Can by the position of the alternatively target of each point in region of search.In order to reduce calculated amount, also can according to probabilistic model Stochastic choice, some put the position of alternatively target in region of search.Also can first based on the position of the target determined in former frame, select in image according to probabilistic model that some put the position of alternatively target at random, then the regions comprising these points are defined as region of search.More than determine that the detail of candidate target and hunting zone is that those skilled in the art can realize, be not described in detail here.

After determining candidate target, can similarity between calculated candidate target and To Template.Similarity is relevant to clarification of objective, and therefore, similarity is actually the similarity compared by the characteristic model of candidate target and To Template.The characteristic model of To Template can obtain in advance.Such as HC feature, HC characteristic model can be obtained based on known To Template.And for HOG feature, the HOG characteristic model that can be obtained by the method for sample training, and based target template need not obtain HOG characteristic model.Especially, if adopt the initial target occurred in HOG detection of classifier video as mentioned above, then can using this HOG sorter as HOG characteristic model.The concrete calculating of similarity is that those skilled in the art can realize, and such as can adopt Euclidean distance method, Pasteur (Bhattacharyya) distance method etc., is not described in detail here.

After calculating the similarity between each candidate target and To Template, just can determine similarity curved surface according to the coordinate of the position of candidate target and the value of corresponding similarity.When the position of some points in region of search described above not by alternatively target, the method for interpolation can be utilized to obtain the value of similarity corresponding to these points.

In step 204, the confidence level of feature is calculated based on the average similarity of relevant similarity curved surface.

In embodiments of the present invention, the confidence level of the feature corresponding to this similarity curved surface is determined above based on the similarity curved surface obtained.Give larger weight for the large feature of confidence level when Fusion Features, thus the position of target can be determined more accurately.

Particularly, the parameter that can reflect the confidence level of feature can be extracted from similarity curved surface.

Extract the parameter of average similarity as the confidence level of reflection feature of similarity curved surface in the present embodiment.Average similarity reflects the average similarity degree between characteristic model and each candidate target, and average similarity degree is higher, and illustrate that this feature more can describe the target of current tracking accurately, namely this feature is more credible.That is, the confidence level of feature becomes positive correlation with average similarity.

In order to reduce the impact of noise, can only be averaged to the similarity that similarity in similarity curved surface is greater than the point of predetermined threshold.Average similarity can be calculated as follows:

$\stackrel{\‾}{d}=\frac{1}{\left|R\right|}\underset{(x,y)\∈R}{\mathrm{\Σ}}D(x,y)$

Wherein, represent average similarity, R represents that in region of search, similarity exceedes the set of the point of threshold value, and (x, y) represents the point in R, and D (x, y) represents the Similarity value at point (x, y) place, | R| represents the element number in set R.Above-mentioned threshold value can be predetermined, also can determine according to the parameter of similarity curved surface.Such as this threshold value can be determined by the value at the top of similarity curved surface, and this threshold value can be 1/2 of the value at top particularly, or the value after the value decay 3dB at top.

Due to average similarity and confidence level positive correlation, can directly adopt average similarity as the confidence level Conf of feature, that is:

$\mathrm{Conf}=\stackrel{\‾}{d}$

Also using the confidence level Conf of the function of average similarity as feature, such as, can be able to define

$\mathrm{Conf}=e^{k\stackrel{\‾}{d}}$

Wherein k be greater than 0 constant, can be such as 3.

In step 206, multiple similarity curved surface carries out synthesizing to obtain synthesis similarity curved surface by the confidence level based on each feature.

Suppose that the confidence level of fisrt feature is Conf ₁, the confidence level of second feature is Conf ₂, then can determine that the weight of the similarity curved surface corresponding to fisrt feature is:

$w_1=\frac{{\mathrm{Conf}}_1}{{\mathrm{Conf}}_1+{\mathrm{Conf}}_2}$

The weight of the similarity curved surface corresponding to second feature is:

$w_2=\frac{{\mathrm{Conf}}_2}{{\mathrm{Conf}}_1+{\mathrm{Conf}}_2}$

So, synthesis similarity curved surface D _cfor the similarity curved surface D of fisrt feature ₁with the similarity curved surface D of second feature ₂weighted sum, namely

D _c＝w ₁D ₁+w ₂D ₂

In a step 208, the position of target is determined based on synthesis similarity curved surface.Such as, the position corresponding to the maximal value in synthesis similarity curved surface can be defined as the position of target in present frame.

By performing above step 202,204,206 and 208 continuously, the position of target can be determined continuously, thus the tracking of realize target.

second embodiment

Fig. 3 shows the schematic diagram of the method for tracking target second embodiment of the invention.Step 302 in Fig. 3,306 and 308 and Fig. 2 in step 202,206 and 208 similar, be not described in detail here.

The difference of the second embodiment and the first embodiment is the step of the confidence level calculating feature.In the step 304 of Fig. 3, the multimodal degree based on relevant similarity curved surface calculates the confidence level of feature.

The schematic diagram of the multi-peaks structure that the similarity curved surface that Fig. 4 shows HOG feature presents.In the process of tracking target, do not having in noisy situation, similarity curved surface often presents single-peak structure.And having interference to deposit in case, similarity curved surface then presents multi-peaks structure, such as HOG feature when two people positions close to often present two peak structure, color histogram feature when have powerful connections interference then easily present multi-peaks structure.Compared to multi-peaks structure, we think that the distribution of single-peak structure is more credible.

The multimodal degree of similarity curved surface can be defined by formula below:

$\mathrm{\γ}=\frac{d_{\mathrm{peak}}^{-}}{d_{\max }}$

Wherein, γ represents the multimodal degree of similarity curved surface, represent the value on the second peak in similarity curved surface, d _maxrepresent the value at the top in similarity curved surface.

In the formula of above-mentioned multimodal degree, the value of γ increases along with the increase of the value on the first peak, and reduce along with the reduction of the value on the second peak, that is γ is less, illustrates that this feature reliability is higher.That is, feature reliability and multimodal degree negative correlation.When similarity curved surface presents single-peak structure, γ is zero.

Because γ is less, this feature reliability is higher, can define confidence level to be:

$\mathrm{Conf}=\frac{1}{\mathrm{\γ}}$

In this definition, γ can not be zero.In order to avoid this situation, another kind of definition can be considered:

Conf＝e ^-cγ

Wherein c be greater than 0 constant, can be such as 0.5.

3rd embodiment

Fig. 5 shows the schematic diagram of the method for the tracking target according to the 3rd embodiment of the present invention.Step 502 in Fig. 5,506 and 508 and Fig. 2 in step 202,206 and 208 similar, be not described in detail here.

The difference of the 3rd embodiment and the first embodiment is the step of the confidence level calculating feature.In the step 504 of Fig. 5, the planarization based on relevant similarity curved surface calculates the confidence level of feature.The ability reflecting feature differentiation target and background of the planarization of similarity curved surface.Curved surface is more smooth, and illustrate that the ability of this feature differentiation object and background is poorer, namely this feature reliability is lower.And curved surface is more precipitous, illustrate that this feature more distinguishes the ability of object and background stronger, namely this feature reliability is higher.That is, the planarization negative correlation of feature reliability and similarity curved surface.

The planarization of similarity curved surface can be defined by formula below:

$S=\frac{1}{\left|P\right|}\underset{(x,y)\∈P}{\mathrm{\Σ}}{I}_{D(x,y)>\mathrm{th}}$

Wherein, S represents planarization, and P represents region of search, | P| represents the number of whole region of search mid point, and (x, y) represents the point in P, and D (x, y) represents the Similarity value at point (x, y) place, th represents threshold value.This threshold value can be predetermined, also can determine according to the parameter of similarity curved surface.Such as this threshold value can be determined by the value at the top of similarity curved surface, and this threshold value can be 1/2 of the value at top particularly, or the value after the value decay 3dB at top.

Because S is less, this feature reliability is higher, can define confidence level to be:

$\mathrm{Conf}=\frac{1}{S}$

4th embodiment

Fig. 6 shows the schematic diagram of the method for the tracking target according to the 4th embodiment of the present invention.Step 602 in Fig. 6,606 and 608 and Fig. 2 in step 202,206 and 208 similar, be not described in detail here.

The difference of the 4th embodiment and the first embodiment is the step of the confidence level calculating feature.In the step 604 of Fig. 6, based on the confidence level of wantonly two or three the calculating features in the average similarity of relevant similarity curved surface, multimodal degree and planarization.

Such as, in following formula confidence level calculating feature can be utilized:

$\mathrm{Conf}=\frac{e^{-\mathrm{c\γ}}{e}^{k\stackrel{\‾}{d}}}{S}$

Or:

$\mathrm{Conf}=\frac{e^{-\mathrm{c\γ}}}{S}$

Or:

$\mathrm{Conf}=\frac{e^{k\stackrel{\‾}{d}}}{S}$

Or:

$\mathrm{Conf}=e^{-\mathrm{c\γ}}{e}^{k\stackrel{\‾}{d}}$

The definition of the symbol in above-mentioned four formula is same as above, no longer repeated description here.

5th embodiment

Fig. 7 shows the schematic diagram of the method for the tracking target according to the 5th embodiment of the present invention.Step 702 in Fig. 7,706 and 708 and Fig. 2 in step 202,206 and 208 similar, be not described in detail here.

The difference of the 5th embodiment and the first embodiment is the step of the confidence level calculating feature.In the step 704 of Fig. 7, calculate the confidence level of feature based on the parameter of relevant similarity curved surface and the parameter of reference similarity curved surface.

Such as, can obtain with reference to similarity curved surface based on initial target location.Particularly, can by centered by initial target location, there is the square as region of search of the predetermined length of side.Or, can will be the center of circle using initial target location, there is the circle of predetermined radii as region of search.The process of all the other processes of computing reference similarity curved surface and the calculating similarity curved surface described in the first embodiment is similar, is not described in detail here.The process of the parameter of computing reference similarity curved surface and reference similarity curved surface can perform after determining initial target location, and can only perform at first once in tracking in tracing process.

Introducing after with reference to similarity curved surface, can revise the formula of the first embodiment to the calculating feature reliability in the 5th embodiment.Particularly, in the 5th embodiment, any one in each feature reliability computing formula in the first to the four embodiment or multiple similarity Surface Parameters can be combined with corresponding reference similarity Surface Parameters value, thus obtain the feature reliability computing formula of correction.

Such as, in following formula confidence level calculating feature can be utilized:

$\mathrm{Conf}=e^{-k(1-\min (1,\stackrel{\‾}{d}/\stackrel{\‾}{d_0}))}$

Or:

$\mathrm{Conf}=e^{c(1-\min (1,\mathrm{\γ}{/\mathrm{\γ}}_0))}$

Or:

$\mathrm{Conf}=\frac{1}{S/S_0}$

Or:

$\mathrm{Conf}=e^{-\mathrm{c\γ}}{e}^{-k(1-\min (1,\stackrel{\‾}{d}/\stackrel{\‾}{d_0}))}$

Or:

$\mathrm{Conf}=e^{c(1-\min (1,\mathrm{\γ}{,\mathrm{\γ}}_0))}{e}^{-k(1-\min (1,\stackrel{\‾}{d}/\stackrel{\‾}{d_0}))}$

Or:

$\mathrm{Conf}=\frac{{e^{-\mathrm{c\γ}}e}^{-k(1-\min (1,\stackrel{\‾}{d}/\stackrel{\‾}{d_0})}}{S}$

Or:

$\mathrm{Conf}=\frac{{e^{c(1-\min (1,\mathrm{\γ}{/\mathrm{\γ}}_0))}e}^{-k(1-\min (1,\stackrel{\‾}{d}/\stackrel{\‾}{d_0})}}{S}$

Or:

$\mathrm{Conf}=\frac{e^{-\mathrm{c\γ}}}{S/S_0}$

Or:

$\mathrm{Conf}=\frac{e^{c(1-\min (1,\mathrm{\γ}/{\mathrm{\γ}}_0))}}{S/S_0}$

Or:

$\mathrm{Conf}=\frac{e^{-k(1-\min (1,\stackrel{\‾}{d}/\stackrel{\‾}{d_0})}}{S/S_0}$

Or:

$\mathrm{Conf}=\frac{e^{-\mathrm{c\γ}}{e}^{-k(1-\min (1,\stackrel{\‾}{d}/\stackrel{\‾}{d_0})}}{S/S_0}$

Or:

$\mathrm{Conf}=\frac{e^{c(1-\min (1,\mathrm{\γ}/{\mathrm{\γ}}_0))}{e}^{-k(1-\min (1,\stackrel{\‾}{d}/\stackrel{\‾}{d_0})}}{S/S_0}$

In above-mentioned formula s ₀, γ ₀represent the average similarity of reference similarity curved surface, planarization and multimodal degree respectively, represent 1 He in less one, represent 1 and γ/γ ₀in less one.The definition of other symbols is same as above, here no longer repeated description.

In above formula, when utilizing average similarity and planarization two parameters at the same time, these two parameters all relate to the selection of the threshold value of similarity.In order to reduce calculated amount, identical threshold value can be selected for these two parameters.Certainly, also different threshold values can be selected.

the device of tracking target

Fig. 8 illustrates the structure of the citing of the computing equipment of the device of the tracking target that may be used for realizing embodiments of the present invention.

In fig. 8, CPU (central processing unit) (CPU) 801 performs various process according to the program stored in ROM (read-only memory) (ROM) 802 or from the program that storage area 808 is loaded into random access memory (RAM) 803.In RAM 803, also store the data required when CPU 801 performs various process etc. as required.

CPU 801, ROM 802 and RAM 803 are connected to each other via bus 804.Input/output interface 805 is also connected to bus 804.

Following parts are connected to input/output interface 805: importation 806, comprise keyboard, mouse etc.; Output 807, comprises display, such as cathode-ray tube (CRT) (CRT) display, liquid crystal display (LCD) etc., and loudspeaker etc.; Storage area 808, comprises hard disk etc.; With communications portion 809, comprise network interface unit such as LAN card, modulator-demodular unit etc.Communications portion 809 is via network such as the Internet executive communication process.

As required, driver 810 is also connected to input/output interface 805.Detachable media 811 such as disk, CD, magneto-optic disk, semiconductor memory etc. are installed on driver 810 as required, and the computer program therefrom read is installed in storage area 808 as required.

Can from network such as the Internet or storage medium such as detachable media 811 to installation procedure computing equipment.

It will be understood by those of skill in the art that this storage medium is not limited to wherein having program stored therein shown in Fig. 8, distributes the detachable media 811 to provide program to user separately with equipment.The example of detachable media 811 comprises disk (comprising floppy disk (registered trademark)), CD (comprising compact disc read-only memory (CD-ROM) and digital universal disc (DVD)), magneto-optic disk (comprising mini-disk (MD) (registered trademark)) and semiconductor memory.Or hard disk that storage medium can be ROM 802, comprise in storage area 808 etc., wherein computer program stored, and user is distributed to together with comprising their equipment.

Below the various embodiments of the device to tracking target of the present invention are described in detail.Wherein, when relating to above in the description of the method for tracking target the aspect related to, for brevity by no longer repeated description.

6th embodiment

Fig. 9 shows the device 900 of the tracking target according to the 6th embodiment of the present invention.Device 900 comprises similarity curved surface computing unit 902, confidence level computing unit 904, similarity curved surface synthesis unit 906 and target location determining unit 908.

Unit in device 900 can be respectively used to execution first embodiment each step to the method for the tracking target of the 5th embodiment.Particularly, similarity curved surface computing unit 902 is configured to calculate multiple similarity curved surfaces relevant to multiple features of target respectively.Confidence level computing unit 904 is configured to the confidence level calculating feature based on the parameter of relevant similarity curved surface.Similarity curved surface synthesis unit 906 is configured to will to state multiple similarity curved surface based on the confidence level of each feature and synthesizes ^withobtain synthesis similarity curved surface.Target location determining unit 908 is configured to the position determining target based on synthesis similarity curved surface.

In one example, confidence level computing unit 904 is configured to the confidence level calculating feature based on the parameter of relevant similarity curved surface and the parameter of reference similarity curved surface.Now, similarity curved surface computing unit 902 can be further configured to computing reference similarity curved surface.

In one example, confidence level computing unit 904 is configured to based on the one or more confidence levels calculating feature in the planarization of relevant similarity curved surface, average similarity and multimodal degree.

The calculating of the design parameter (such as planarization, average similarity and multimodal degree) of similarity curved surface and/or reference similarity curved surface can be performed by similarity curved surface computing unit 902, also can be performed by confidence level computing unit 904.

Further details about the operation of each unit of device 900 with reference to the description of the first embodiment to the 5th embodiment, no longer can repeat here.

Above some embodiments of the present invention are described in detail.As one of ordinary skill in the art can be understood, whole or any step of method and apparatus of the present invention or parts, can in the network of any computing equipment (comprising processor, storage medium etc.) or computing equipment, realized with hardware, firmware, software or their combination, this is that those of ordinary skill in the art use their basic programming skill just can realize when understanding content of the present invention, therefore need not illustrate at this.

In addition, it is evident that, when relating to possible peripheral operation in superincumbent explanation, any display device and any input equipment, corresponding interface and control program that are connected to any computing equipment will be used undoubtedly.Generally speaking, related hardware in computing machine, computer system or computer network, software and realize the hardware of the various operations in preceding method of the present invention, firmware, software or their combination, namely form equipment of the present invention and each building block thereof.

Therefore, based on above-mentioned understanding, object of the present invention can also be realized by an operation program or batch processing on any messaging device.Described messaging device can be known common apparatus.Therefore, object of the present invention also can realize only by the program product of providing package containing the program code realizing described method or equipment.That is, such program product also forms the present invention, and stores or the medium that transmits such program product also forms the present invention.Obviously, described storage or transmission medium can be well known by persons skilled in the art, or the storage of any type developed in the future or transmission medium, therefore also there is no need to enumerate various storage or transmission medium at this.

Present invention also offers a kind of computer program, for realizing following methods: calculate multiple similarity curved surfaces relevant to multiple features of target respectively, this similarity curved surface is the curved surface that one group of similarity between candidate target and To Template is formed; Parameter based on relevant similarity curved surface calculates the confidence level of feature; Multiple similarity curved surface carries out synthesizing to obtain synthesis similarity curved surface by the confidence level based on each feature; And the position of target is determined based on synthesis similarity curved surface.

Present invention also offers a kind of computer program of at least computer-readable medium form, it recording the computer program code of the method for realizing above-mentioned tracking target.

In equipment of the present invention and method, obviously, each parts or each step reconfigure after can decomposing, combine and/or decomposing.These decompose and/or reconfigure and should be considered as equivalents of the present invention.Also it is pointed out that the step performing above-mentioned series of processes can order naturally following the instructions perform in chronological order, but do not need necessarily to perform according to time sequencing.Some step can walk abreast or perform independently of one another.Simultaneously, in specific description of embodiments of the present invention above, the feature described for a kind of embodiment and/or illustrate can use in one or more other embodiment in same or similar mode, combined with the feature in other embodiment, or substitute the feature in other embodiment.

Should emphasize, term " comprises/comprises " existence referring to feature, key element, step or assembly when using herein, but does not get rid of the existence or additional of one or more further feature, key element, step or assembly.

Although described the present invention and advantage thereof in detail, be to be understood that and can have carried out various change when not exceeding the spirit and scope of the present invention limited by appended claim, substituting and conversion.And the scope of the application is not limited only to the embodiment of process, equipment, means, method and step described by instructions.One of ordinary skilled in the art will readily appreciate that from disclosure of the present invention, can use perform the function substantially identical with corresponding embodiment described herein or obtain and its substantially identical result, existing and that will be developed in the future process, equipment, means, method or step according to the present invention.Therefore, appended claim is intended to comprise such process, equipment, means, method or step in their scope.

The present invention can be applicable to the fields such as video monitoring, for monitoring pedestrian, animal or other targets.The present invention, by assessing the confidence level of each road feature more accurately, can realize better Fusion Features effect, thus can more effectively tracking target.

Claims (4)

1. a method for tracking target, comprising:

Calculate multiple similarity curved surfaces relevant to multiple features of target respectively, described similarity curved surface is the curved surface that one group of similarity between candidate target and To Template is formed;

The confidence level of feature is calculated based on the parameter of relevant similarity curved surface and the parameter of reference similarity curved surface, wherein said with reference to similarity curved surface be described target be in initial position time the corresponding similarity curved surface that obtains, the parameter of described relevant similarity curved surface and the confidence level of the described parameter with reference to similarity curved surface reflection individual features;

Described multiple similarity curved surface is weighted addition to obtain synthesis similarity curved surface by the confidence level based on each described feature, and the feature that wherein confidence level is large is endowed larger weight; And

The position of described target is determined based on described synthesis similarity curved surface,

The confidence level wherein calculating feature comprises the planarization based on relevant similarity curved surface, one or more and the described planarization with reference to similarity curved surface in average similarity and multimodal degree, one or more confidence levels calculating feature in average similarity and multimodal degree, the number that the planarization of wherein said relevant similarity curved surface and described reference similarity curved surface is greater than the point of predetermined threshold based on similarity in corresponding similarity curved surface is respectively determined, described relevant similarity curved surface and the described average similarity with reference to similarity curved surface are the average similarity that in corresponding similarity curved surface, similarity is greater than the point of predetermined threshold respectively, described relevant similarity curved surface and the described multimodal degree with reference to similarity curved surface are the values on the second peak in corresponding similarity curved surface and the ratio of the value at top respectively,

The step of the confidence level of wherein said calculating feature comprises the confidence level being calculated feature by following formula:

$\mathrm{Conf}=\frac{e^{-\mathrm{c\γ}}{e}^{-k(1-\min (1,\stackrel{\‾}{d}/\stackrel{\‾}{d_0}))}}{S/S_0}$

The wherein confidence level of Conf representation feature, represent the average similarity of relevant similarity curved surface, represent the average similarity with reference to similarity curved surface, S represents the planarization of relevant similarity curved surface, S ₀represent the planarization with reference to similarity curved surface, γ represents the multimodal degree of relevant similarity curved surface, c, k be more than or equal to 0 constant, represent 1 He in less one.

2. the method for claim 1, wherein said multiple feature comprises at least one global characteristics and at least one local feature.

3. method as claimed in claim 2, wherein said global characteristics is color histogram, and described local feature is selected from: gradient orientation histogram, Haar wavelet character, edge feature, local binary patterns feature.

4. a device for tracking target, comprising:

Similarity curved surface computing unit, is configured to calculate multiple similarity curved surfaces relevant to multiple features of target respectively, and described similarity curved surface is the curved surface that one group of similarity between candidate target and To Template is formed;

Confidence level computing unit, be configured to the confidence level calculating feature based on the parameter of relevant similarity curved surface and the parameter of reference similarity curved surface, wherein said with reference to similarity curved surface be described target be in initial position time the corresponding similarity curved surface that obtains, the parameter of described relevant similarity curved surface and the confidence level of the described parameter with reference to similarity curved surface reflection individual features;

Similarity curved surface synthesis unit, described multiple similarity curved surface is weighted addition to obtain synthesis similarity curved surface by the confidence level be configured to based on each described feature, and the feature that wherein confidence level is large is endowed larger weight; And

Target location determining unit, is configured to the position determining described target based on described synthesis similarity curved surface,

Wherein said confidence level computing unit is configured to the planarization based on relevant similarity curved surface, one or more and the described planarization with reference to similarity curved surface in average similarity and multimodal degree, one or more confidence levels calculating feature in average similarity and multimodal degree, the number that the planarization of wherein said relevant similarity curved surface and described reference similarity curved surface is greater than the point of predetermined threshold based on similarity in corresponding similarity curved surface is respectively determined, described relevant similarity curved surface and the described average similarity with reference to similarity curved surface are the average similarity that in corresponding similarity curved surface, similarity is greater than the point of predetermined threshold respectively, described relevant similarity curved surface and the described multimodal degree with reference to similarity curved surface are the values on the second peak in corresponding similarity curved surface and the ratio of the value at top respectively,

Wherein said confidence level computing unit is configured to the confidence level being calculated feature by following formula:

$\mathrm{Conf}=\frac{e^{-\mathrm{c\γ}}{e}^{-k(1-\min (1,\stackrel{\‾}{d}/\stackrel{\‾}{d_0}))}}{S/S_0}$

The wherein confidence level of Conf representation feature, represent the average similarity of relevant similarity curved surface, represent the average similarity with reference to similarity curved surface, S represents the planarization of relevant similarity curved surface, S ₀represent the planarization with reference to similarity curved surface, γ represents the multimodal degree of relevant similarity curved surface, c, k be more than or equal to 0 constant, represent 1 He in less one.