CN101246546A - Variable shelter template matching algorithm of video target tracking - Google Patents

Abstract

The invention specifically relates to a variable masking template matching algorithm in video target tracking which belongs to the field of computer vision and pattern recognition technology. In video target tracking, the target is obscured often by other object partly. In such circumstances, using the traditional fixed masking template matching to determine target location will cause accuracy decreased significantly. In order to realize target location accurately under a condition that the target is obscured partly, the invention combines with block analysis technology, on the basis of the first non-exact match result using variable masking template matching algorithm to do correction, the template masking in the process of correction changes dynamically as the change of candidate target location, thus causing non-masking part of the target always can guide the matching algorithm effectively to find precise target location. Confirms effectiveness of the invention algorithm based on experimental result of real video streaming.

Description

Variable shelter template matching algorithm in a kind of video frequency object tracking

Technical field

The invention belongs to computer vision and pattern analysis technical field, be specifically related to the variable shelter template matching algorithm in a kind of video frequency object tracking.

Background technology

Target following has a wide range of applications in man-machine interaction, supervision automatically, video frequency searching, Traffic monitoring and automobile navigation.The task of target following is to determine how much states of target in each frame of video flowing, comprises position, size and orientation etc.Owing to do not limit the outward appearance of tracked target, and the outward appearance of target can change in tracing process, adds the interference of complicated prospect and background, and target tracking algorism is faced with lot of challenges, is one of research focus of computer vision field.

Target tracking algorism can be divided three classes, and a class is a tracking (point tracking) ^[1,2], second class is that nuclear is followed the tracks of (kerneltracking) ^[3～7,10], the 3rd class is that silhouette is followed the tracks of (silhouette tracking) ^[8,9]The target tracking algorism that the present invention proposes belongs to the nuclear track algorithm.This algorithm characterizes target with display model (that is template), and the geological information of target in each frame described with affine transformation parameter usually ^[10]

For the nuclear track algorithm, one of maximum challenge is exactly processing target this problem that is blocked how ^[3,5-7]Why this problem is difficult to solve is because target and shelter can be any outward appearances, and the time of blocking also can be arbitrarily.Document [11] has provided a kind of effective sheltering analysis algorithm, can analyze the situation that current goal is blocked under the situation of the exact position that obtains target effectively, shelters so that carry out the used template of template matches in the generation next frame.Yet, also have the problem of a key not solve: because the situation of blocking of target is in continuous variation, the target occlusion situation of present frame is sheltered in many cases and be not suitable for to the template that obtains according to the situation of blocking of target in the previous frame, thereby shelter with this template and to carry out the error that template matches can cause target localization, especially block the serious or change situation of blocking of comparison relatively significantly the time when target, this error can be very big, even cause and follow the tracks of failure.

Will realize under unknown present frame target occlusion situation that in fact the precision target location needs to solve an antinomy of prerequisite each other: accurately localizing objects need at first be known the target occlusion situation of present frame, and the target occlusion situation of present frame only can be by comparing to determine with To Template behind the exact position that has obtained target.Up to the present, also there is not document to propose to solve the method for this antinomy.

List of references

[1]C.Rasmussen，and G.Hager.Probabilistic data association methods for tracking complexvisual objects.IEEE Trans.on Pattern Analysis and Machine Intelligence，23(6)：560-576，2001.

[2]C.Hue，J.L.Cadre，P.Prez.Sequential Monte Carlo methods for multiple target tracking anddata fusion.IEEE Trans.on Signal Processing，50(2)：309-325，2002.

[3]A.D.Jepson，D.J.Fleet，and T.F.EI-Maraghi.Robust online appearance model for visualtracking.IEEE.Trans.on Pattern Analysis and Machine Intelligence，25(10)：1296-1311，2003.

[4]D.Comaniciu，V.Ramesh，and P.Meer.Kernel-based object tracking，IEEE Trans.onPattern Analysis and Machine Intelligence，25(5)：564-577，2003

[5]S.K.Zhou，R.Chellappa，and B.Moghaddam.Visual tracking and recognition usingappearance-adaptive models for particle filters.IEEE Trans.on Image Processing，13(11)：1491-1506，2004.

[6]H.T.Nguyen，M.Worring，and R.van den Boomgaard.Occlusion robust adaptive templatetracking.Proc.IEEE Int’l Conf.Computer Vision，1：678-683，2001.

[7]H.T.Nguyen，and A.W.M.Smeulders.Fast occluded object tracking by a robust appearancefilter.IEEE Trans.on Pattern Analysis and Machine Intelligence，26(8)：1099-1104，2004.

[8]Y.Chen，Y.Rui，and T.Huang.Jpdaf based HMM for real-time contour tracking.Proc.IEEE Conf.on Computer Vision and Pattern Recognition，1：543-550，2001.

[9]A.Yilmaz，X.Li，and M.Shan.Contour based object tracking with occlusion handling invideo acquired using mobile cameras.IEEE Trans.on Pattern Analysis and MachineIntelligence，26(11)：1531-1536，2004.

[10]S.Baker，and I.Matthews.Lucas-Kanade 20years on：a unifying framework.Int’l JournalComputer Vision，53(3)：221-255，2004.

[11] Pan Jiyan, Hu Bo, Zhang Jianqiu, " a kind of content self-adaptive gradual-progression type sheltering analysis target tracking algorism ", number of patent application: 200710045941.7.

Summary of the invention

The objective of the invention is to propose the variable shelter template matching algorithm in a kind of video frequency object tracking, realize accurate target localization under by the situation of partial occlusion in order to solve in target.

How key of the present invention is to utilize sheltering analysis behind the non-exact position that obtains target ^[11]The result, design a kind of new template matching algorithm, the position of correction target.

Because variation may take place the situation of blocking of target in the present frame, the template that the situation of blocking by former frame generates is sheltered the template matching algorithm that often can't make in the present frame and is found accurate target location.Thereby, in present frame, carry out the fixed mask template matches after, resulting target area (is the area-of-interest in the document [11], ROI) often with the position at the real place of target some deviations is arranged.The part target can be in outside the ROI.This situation is shown in the subgraph in the lower left corner among Fig. 1.Thereby the interference figure U that utilizes the sheltering analysis algorithm of document [11] to obtain can produce mistake.But the sheltering analysis result who is positioned at the image within the ROI is still reliable.Therefore can utilize this part information to mate to come the position of correction target by variable shelter template.

Therefore, method of the present invention is, obtains the non-exact position of target by the fixed mask template matches, and the template when changing template matches according to the result of sheltering analysis is then sheltered, it is accurately located under the situation of partial occlusion to be implemented in target, thereby obtained the exact position of target.

The coupling of variable shelter template coupling and way of search and conventional fixed mask template matches ^[5-7]Identical, but to be sequestered in the search procedure be not to be changeless to its template, but change along with the variation of candidate's coordinate conversion parameter.Specifically, variable template is sheltered according to candidate's coordinate transform and is obtained by interference figure U is sampled:

M _A(x；a)＝1-U{round[φ(x；a)]} (1)

M wherein _A(x; A) be variable template value at template coordinate points x when being sequestered in template matches candidate coordinate conversion parameter and being a; φ (x; A) expression is that the coordinate transform φ of a is mapped to template coordinate points x in the frame of video coordinate by transformation parameter; The round operational character is represented to round; U is the interference figure that obtains through sheltering analysis after the fixed mask template matches, and its pixel can only value 0 or 1, so M _AValue also can only be 0 or 1.It is not impact point that certain pixel of U is got 1 this place of expression, M _ACertain pixel get 1 the expression this point masked when template matches.In Fig. 1, U and M _AWhite portion show that value is 1, the black part branch shows that value is 0.

The variable template of (1) formula of employing definition is sheltered, and conventional fixed mask template matches just changes variable shelter template into and mated:

$a_A=\underset{a}{\arg \min }\frac{1}{\mathrm{sum}\left(M_A\right)}\mathrm{sim}\left\{I\right[\mathrm{\φ}\left(a\right)]\⊗M_A\left(a\right),T\⊗M_A\left(a\right)\}---\left(2\right)$

A wherein _AIt is coordinate conversion parameter through the reflection present frame target exact position that obtains after the variable shelter template coupling; Sim{A, any one measuring similarity between B} presentation video A and the B; I[φ (a)] be to be that the coordinate transform φ of a is mapped to resulting partial video two field picture in the frame of video coordinate with template by transformation parameter; M _A(a) be that whole variable template when template matches candidate coordinate conversion parameter is a is sheltered; T is the current target template; The  operator representation multiplies each other the pixel value of corresponding point between two width of cloth images; Sum (M _A) represent M _AThe summation of all pixel values.

Fig. 1 has provided the synoptic diagram of the variable shelter template coupling of (2) formula definition.Tracked target is marked with letter " A ", and " B ", " C ", " D " is to distinguish the zones of different of target.In present frame, the ROI with dashed lines frame table that the fixed mask template matches by routine obtains shows that the ROI that obtains by the variable shelter template coupling represents with solid box.Noticing that variable template is sheltered along with the change of candidate target region changes, thereby has produced variable template after the processing and the measuring similarity between the target area sheltered of different processes.As seen from Figure 1, variable shelter template coupling why can the correction target position, be because in the coordinate conversion parameter search procedure of (2) formula, when real target and candidate target region are not overlapping, not masked template is always inequality with not masked candidate target region, and only just can obtain maximal value at precise coordinates transformation parameter place through variable image similarity tolerance of sheltering after the processing, thereby (2) formula of making searches accurate target location.

Description of drawings

Fig. 1: variable shelter template coupling synoptic diagram.

Fig. 2: tracking performance comparative example before and after variable shelter template matching algorithm adds.Tracking results is shown with the white rectangle frame table of center band cross.(a ₁)-(a ₄): when not adding variable shelter template matching algorithm, the track algorithm of document [11] is with losing target; (b ₁)-(b ₄): after adding variable shelter template matching algorithm, the track algorithm of document [11] has been caught up with target well; (c ₁)-(c ₄): when not adding variable shelter template matching algorithm, the tracking accuracy of the track algorithm of document [11] is lower; (d ₁)-(d ₄): after adding variable shelter template matching algorithm, the track algorithm of document [11] has been obtained higher tracking accuracy.In first and second row, the image of demonstration is taken from the 765th, 877,983 of video flowing, with 1004 frames; The 3rd with fourth line in, the image of demonstration is taken from the 1202nd, 1295,1344 of video flowing, with 1399 frames.

Embodiment

In concrete enforcement of the present invention, the track algorithm that adopts document [11] proposition relatively adds the tracking performance of the variable shelter template matching algorithm front and back that the present invention proposes as the basis.

We have at first done above-mentioned comparison on a large amount of outdoor scene video flowings.These outdoor scene video flowings comprise different types of target and the various scene of blocking, and in addition, the motion of camera is arbitrarily.We are divided into two types to the scene of blocking of 30 test video streams: short-term is blocked and is blocked for a long time.If a duration of blocking surpasses 25 frames, then be considered to block for a long time.Experimental result is as shown in table 1.

Table 1

Type of barrier	Document [11]	The present invention
			Short-term (15)	14	15
(15) for a long time	13	14
			Amount to (30)	27	29

By table 1 as seen, add variable shelter template matching algorithm after, tracking performance has had further raising.

Two typical examples in the above-mentioned outdoor scene video flowing as shown in Figure 2.As seen from Figure 2, add variable shelter template matching algorithm after, the stability and the degree of accuracy of tracking all significantly improve.

Claims (3)

1, the variable shelter template matching algorithm in a kind of video frequency object tracking, it is characterized in that obtaining the non-exact position of target by the fixed mask template matches, template when changing template matches according to the result of sheltering analysis is then sheltered, it is accurately located under the situation of partial occlusion to be implemented in target, thereby obtained the exact position of target.

2, the variable shelter template matching algorithm in the video frequency object tracking according to claim 1 is characterized in that the mode that variable template shelters is as follows:

M _A(x；a)＝1-U{round[φ(x；a)]}

M wherein _A(x; A) be variable template value at template coordinate points x when being sequestered in template matches candidate coordinate conversion parameter and being a; φ (x; A) expression is that the coordinate transform φ of a is mapped to template coordinate points x in the frame of video coordinate by transformation parameter; The round operational character is represented to round; U is the interference figure that obtains through sheltering analysis after the fixed mask template matches.

3, the variable shelter template matching algorithm in the video frequency object tracking according to claim 1 and 2, it is as follows to it is characterized in that utilizing variable template to shelter the mode of carrying out template matches:

$a_A=\underset{a}{\arg \min }\frac{1}{\mathrm{sum}\left(M_A\right)}\mathrm{sim}\left\{I\right[\mathrm{\φ}\left(a\right)]\⊗M_A\left(a\right),T\⊗M_A\left(a\right)\}$

A wherein _AIt is coordinate conversion parameter through the reflection present frame target exact position that obtains after the variable shelter template coupling; Sim{A, any one measuring similarity between B} presentation video A and the B; I[φ (a)] be to be that the coordinate transform φ of a is mapped to resulting partial video two field picture in the frame of video coordinate with template by transformation parameter; M _A(a) be that whole variable template when template matches candidate coordinate conversion parameter is a is sheltered; T is the current target template; The  operator representation multiplies each other the pixel value of corresponding point between two width of cloth images; Sum (M _A) represent M _AThe summation of all pixel values.

Images