CN101564300B - Gait cycle detection method based on regional characteristics analysis - Google Patents

Abstract

The invention provides a gait cycle detection method based on regional characteristics analysis and the method includes the acquisition of the objective outline of a pedestrian and detection of gait cycle. The acquisition method of the objective outline includes the following steps of: first extracting a single frame image for grey scale transformation from a video, then calculating the median of all pixel points frame by frame as a background image of the whole sequence, and finally adopting a background subtraction to extract a human objective, using mathematical morphology to fill the cavity of a binary image, analyzing and extracting the profile of human by single connection, placing the human body in the middle and unifying the dimensions of images to be 64*64pixel; and the gait cycledetection is to convert the gait cycle analysis problem to an image regional characteristics analysis problem with a single frame, namely analyzing the gait cycle according to the characteristic chan ge situation of image regions in each frame. The method not only has small calculation quantity, but also already reaches the precision to which people subjectively judge the gait cycle, and avails the possibility of real-time gait identification.

Description

Gait cycle detecting method based on regional characteristics analysis

(1) technical field

What the present invention relates to is a kind of mode identification technology, the gait cycle detecting method in specifically a kind of Gait Recognition.

(2) background technology

The purpose of Gait Recognition is to carry out identification according to the posture that people walk.Gait Recognition can obtain this gait feature under the situation that does not allow object of study perceive, have non-infringement, untouchable, less demanding, remote to systemic resolution, be difficult to camouflage, advantage such as little affected by environment.Therefore, from the viewpoint of video monitoring, gait is the most potential biological characteristic under the remote situation.Gait Recognition is with a wide range of applications and economic worth in fields such as gate control system, security monitoring, man-machine interaction, medical diagnosiss, has therefore excited the research enthusiasm of domestic and international vast researcher.

The gait sequence image is periodic space-time unite signal, discerns individuality if study whole section video, not only causes the big shortcoming of data volume, and also exists redundant on the information.Therefore need the method for utilization cycle analysis to determine start frame and end frame, and then determine a gait cycle, thereby in one-period, extract feature, reach the purpose of final identification.The home and abroad Many researchers is studied on gait cycle detects, and people such as BenAbdelkader determine gait cycle by the self-similarity that calculates human body contour outline; People such as BenAbdelkader are also according to the periodicity of the width analysis gait of border rectangle frame; People such as Collins have analyzed the cyclically-varying of human body height and width, and then the observation gait cycle; People such as Kale analyze the cyclophysis of gait over time by the norm of observing human body width vector; The autocorrelation of human foreground pixel sums such as Boulgouris is judged the cycle of gait; People such as Sarkar adopt what cyclophysis of human region the latter half pixel to determine the cyclically-varying of gait; People such as Li line up self similarity figure (SSP) with gait, adopt the linear local Nonlinear Dimension Reduction method that embeds (LLE) to extract the periodicity that the unidimensional feature that has kept original geometry is analyzed gait then; People such as Chen Shi with all pedestrian's contour area boundary rectangle frames in the gait sequence as image-region, image-region the end of from and on 1/4 height in, the equivalent level is cut apart three zones, calculates each district's accumulative total profile and counts, and is put the distribution histogram feature detection accordingly and goes out gait cycle; The big shortcoming of these method ubiquity amounts of calculation.Because the order of accuarcy that gait cycle is cut apart has had a strong impact on the precision of Gait Recognition problem, existing most of documents all are to cut apart the Gait Recognition algorithm that proposes under the good situation at the supposition gait cycle.

Open report related to the present invention comprises:

[1]BenAbdelkader?C，Culter?R，Davis?L.Motion?based?recognition?of?people?in?eigengaitspace[C]In：proceedings?of?the?IEEE?International?Conference?on?Automatic?Face?and?GestureRecognition.Washington?DC，USA，2002：254-259P；

[2]Boulgouris?N?V，Plataniotis?K?N，Hatzinakos?D.Gait?recognition?using?dynamic?timewarping[C].2004?IEEE?6th?Workshop?on?Multimedia?Signal?Processing，2004：263-266；

[3]Li?Hong-gui，Shi?Cui-ping，Li?Xing-guo.LLE?based?gait?recognition[C].In：Proceedings?of2005?International?Conference?on?Machine?Learning?and?Cybernetics，2005，7：4516-4521；

[4] old reality, Ma Tianjun, Huang Wanhong, etc. be used for the multilamellar video in window square of Gait Recognition. electronics and information journal, 2009,31 (1): 116-119.

(3) summary of the invention

The object of the present invention is to provide and a kind ofly can effectively improve gait cycle detection speed and precision, thereby provide possible gait cycle detecting method for real-time Gait Recognition based on regional characteristics analysis.

The object of the present invention is achieved like this:

Comprise that obtaining with gait cycle of pedestrian's objective contour detect; The method of obtaining of described pedestrian's objective contour is: at first extract single-frame images and carry out greyscale transformation from video, calculate the intermediate value of each pixel in frame by frame then, background image as whole sequence, adopt the background subtraction method to extract human body target at last, fill up cavity, the simply connected analysis of binary image with mathematical morphology and extract people's silhouette, make human body placed in the middle, the size of image is unified to be the 64*64 pixel; It is the graphics field feature analysis problem that the gait cycle problem analysis is converted into single frames that described gait cycle detects, promptly analyze the cycle of gait, be a gait cycle to occurring local extremum for the third time again from local extremum occurring for the first time according to the changing features situation of graphics field in every frame.

The present invention can also comprise:

1, described graphics field feature is a kind of in area, barycenter, fitted ellipse or circle, particular point, the bounding box feature of graphics field in every frame.

2, described area be area behind intrinsic area, convex hull area, the filling cavity or intrinsic area account for convex hull ratio in a kind of.

3, described fitted ellipse or circle are the long and short axle of the ellipse of human region with identical second order spatial moment and eccentricity or the circle that equates with the area of human region.

4, the described bounding box feature variation that width changes frame by frame or movement human accounts for the bounding box ratio that is the movement human bounding box.

Main effect of the present invention is: not only amount of calculation is little, and has reached the precision of people's subjective judgment gait cycle, for real-time Gait Recognition provides possibility.

(4) description of drawings

The flow chart that Fig. 1 gait cycle detects;

Fig. 2 (a)-Fig. 2 (e) extracts the preprocessing process of human body target, wherein: Fig. 2 (a) greyscale transformation, the reconstruction of Fig. 2 (b) background, Fig. 2 (c) background subtraction, Fig. 2 (d) human body contour outline, Fig. 2 (e) standardization centralization;

Fig. 3 gait sequence image;

Fig. 4 convex hull sketch map;

Fig. 5 (1)-Fig. 5 (20) the whole bag of tricks detects gait cycle, wherein Fig. 5 (1) is according to intrinsic area, Fig. 5 (2) is according to the convex hull area, Fig. 5 (3) is according to the area behind the filling cavity, Fig. 5 (4) accounts for the ratio of convex hull according to figure, Fig. 5 (5) is according to the vertical coordinate of barycenter, Fig. 5 (6) is according to the abscissa of barycenter, Fig. 5 (7) is according to the fitted ellipse major axis, Fig. 5 (8) is according to the fitted ellipse minor axis, and Fig. 5 (9) is according to the fitted ellipse eccentricity, and Fig. 5 (10) is according to the match diameter of a circle, Fig. 5 (11) is according to the right-top abscissa, Fig. 5 (12) is according to the right-top vertical coordinate, and Fig. 5 (13) is according to the right-bottom abscissa, and Fig. 5 (14) is according to the right-bottom vertical coordinate, Fig. 5 (15) is according to the left-bottom abscissa, Fig. 5 (16) is according to the left-bottom vertical coordinate, and Fig. 5 (17) is according to the left-top abscissa, and Fig. 5 (18) is according to the left-top vertical coordinate, Fig. 5 (19) is according to the width of human body bounding box, and Fig. 5 (20) accounts for the ratio of bounding box according to figure;

There is the gait single-frame images in cavity in Fig. 6, and 4 frames among Fig. 6 are respectively the 10th among Fig. 3,12,22 and 49 frames from left to right, and circle has marked the cavity that occurs on the true gait profile;

The ellipse of Fig. 7 graphics field match, wherein 1 is oval focus, and 2 is long axis of ellipse, and 3 is oval minor axis, 4 ellipses;

The location of Fig. 8 (1)-(2) eight particular points of Fig. 8, Fig. 8 (1) is an ordinary circumstance, Fig. 8 (2) is the special circumstances of (1), and wherein 1 is top-right, and 2 is right-top, 3 is right-bottom, 4 is bottom-right, and 5 is top-left, and 6 is left-top, 7 is left-bottom, and 8 is bottom-left;

The bounding box of Fig. 9 human region.

(5) specific embodiment

Gait cycle detecting method based on regional characteristics analysis of the present invention comprises that obtaining with gait cycle of pedestrian's objective contour detect; The method of obtaining of described pedestrian's objective contour is: at first extract single-frame images and carry out greyscale transformation from video, calculate the intermediate value of each pixel in frame by frame then, background image as whole sequence, adopt the background subtraction method to extract human body target at last, fill up cavity, the simply connected analysis of binary image with mathematical morphology and extract people's silhouette, make human body placed in the middle, the size of image is unified to be the 64*64 pixel; It is the graphics field feature analysis problem that the gait cycle problem analysis is converted into single frames that described gait cycle detects.Promptly analyze the cycle of gait according to the changing features situation of graphics field in every frame; Be a gait cycle from local extremum occurring for the first time to occurring local extremum for the third time again.

Described gait cycle detecting method is: the cycle of analyzing gait according to the situation of change of the features such as area, barycenter, fitted ellipse or circle, particular point and bounding box of graphics field in every frame.

The cycle that described area change situation according to graphics field in every frame is analyzed gait is: the ratio that accounts for convex hull according to the area behind the intrinsic area of movement human, convex hull area, the filling cavity and intrinsic area is carried out gait cycle and is detected, and is a gait cycle from local extremum occurring for the first time to occurring local extremum for the third time again.

The cycle that described barycenter situation of change according to graphics field in every frame is analyzed gait is: according to the barycenter vertical coordinate of movement human and abscissa frame by frame situation of change carry out gait cycle and detect, be a gait cycle from local extremum occurring for the first time to occurring local extremum for the third time again.

The cycle that described fitted ellipse according to graphics field in every frame (or circle) situation of change is analyzed gait is: according to the diameter of a circle that has the long and short axle of ellipse of identical second order spatial moment and eccentricity with human region or equate with the area of human region frame by frame situation of change carry out gait cycle and detect, local extremum occurs from the first time and be a gait cycle to occurring local extremum for the third time again;

Described particular point situation of change according to graphics field in every frame is analyzed the cycle of gait: because in the process that the people walks, shoulder rocks back and forth with respect to mass center of human body, adopt the right-top point to judge that the cyclically-varying of gait is effective, be a gait cycle to occurring local extremum for the third time again from local extremum occurring for the first time;

The cycle that described bounding box situation of change according to graphics field in every frame is analyzed gait is: change frame by frame or carry out gait cycle according to the variation that movement human accounts for the bounding box ratio and detect according to the width of movement human bounding box, be a gait cycle from local extremum occurring for the first time to occurring local extremum for the third time again.

For example the present invention is done description in more detail below in conjunction with accompanying drawing:

1. in order to extract human body target, at first from original video, extract single-frame images and carry out greyscale transformation (as Fig. 2 (a)); Calculate then each pixel in frame by frame intermediate value, as the background image (as Fig. 2 (b)) of whole sequence; At last, adopt the background subtraction method to extract human body target (as Fig. 2 (c)), fill up cavity, the simply connected analysis of binary image with mathematical morphology and extract people's silhouette (as Fig. 2 (d)). for the removal of images size should make human body placed in the middle to the influence of discerning, the size of image is unified to be 64*64 pixel (as Fig. 2 (e)).

2. gait cycle detects

After carrying out foregoing image sequence pretreatment, the gait cycle problem analysis is converted into the graphics field feature analysis problem of single frames.With one section sequence (as Fig. 3) that contains 53 frames is the gait cycle that example is analyzed it.The 10 kinds of gait cycle detecting methods based on regional characteristics analysis of five classes that analyze to propose are respectively that attribute, some particular points of area, barycenter, fitted ellipse (or circle) according to movement human in the image, the situation of change of bounding box are judged the cycle of gait.

2.1 the gait cycle detecting method of----area based on provincial characteristics

Here, for the graphics field area features, we mainly consider is area behind intrinsic area, convex hull area and the filling cavity of movement human.

Making the set of pixels in the zone is R, if f (r, c)=1, A represents the intrinsic area in zone:

$A=\underset{(r,c)\∈R}{\mathrm{\Σ}}f(r,c)---\left(1\right)$

F (r, c) expression (r, the gray value of c) locating.A means the number of image-region pixel.

In computational geometry, the convex hull of the point set X among the real vector space V is the minimum border that comprises this formed zone of point set, and this point set is the limited convex set of non-NULL on the plane, is illustrated in figure 4 as the intuition picture of a convex hull.For a planar object, convex hull is easy to be thought of as the polygon wire frame of the encirclement certain objects that elasticity extends, and the ductile polygon wire frame of this elasticity of imagination is that one deck is subjected to tensile film.But balanced surface (least energy face) may not be a convex hull in this case.For the convex hull of the set X that illustrates at real vector space V exists, more directly express the form of convex set, make the convex set of X be described to come from the convex set of the finite subset of X mid point, the form of the point in the set is a ∑ _J=1 ⁿt _jx _j, n is any natural number here, t _jBe nonnegative number, and ${\mathrm{\Σ}}_{j=1}^n{t}_j=1,$ x _j(j=1 ..., n) be point among the X.So the convex set of X is expressed as H _Convex(X):

$H_{\mathrm{convex}}\left(X\right)=\left\{\underset{i=1}{\overset{k}{\mathrm{\Σ}}}{\mathrm{\α}}_i{x}_i\right|x_i\∈X,{\mathrm{\α}}_i\∈R,{\mathrm{\α}}_i\≥0,\underset{i=1}{\overset{k}{\mathrm{\Σ}}}{\mathrm{\α}}_i=1,k=\mathrm{1,2},\·\·\·\}---\left(2\right)$

(for example: whole SPACE V), the common factor of the convex set of any X of comprising is all common factors that comprise the convex set of X to note comprising a convex set at least by X.In fact, if X is the subclass of a N gt, use N+1 salient point in conjunction with just enough at most according to top definition.Two dimensional surface and three dimensions finite point set and geometric object are special cases.

2.1.1 observe the periodicity of gait according to the intrinsic area change of movement human in every two field picture

Shown in Fig. 5 (1), when the open angle of human limb is maximum (as the 5th in the image sequence, 17,29 and 41 frames), bianry image white portion maximum, the area maximum of movement human in the bianry image this moment; When extremity close up (as the 11st in the image sequence, 23,35 and 47 frames), the area minimum of movement human this moment.Therefore the area by movement human in the single-frame images is the periodic of gait sequence image as can be seen.Be exactly a gait cycle for example from the 5th frame to the 28 frames; From the 11st frame to the 34 frames also is a complete cycle.In this section gait sequence, can find a lot of groups of gait cycles, because the not definition of the starting point in cycle.Especially, we can define: the starting point of each gait cycle is the open angle maximums of extremity, is half period from this frame when occurring for the second time that the open angle of extremity is maximum; Is the whole cycle from this frame when occurring for the third time that the open angle of extremity is maximum.

2.1.2 observe the periodicity of gait according to the cyclically-varying of the convex hull area in movement human zone in every two field picture

Use periodicity that how many numbers of pixel in the convex hull survey gait shown in Fig. 5 (2), but when if the influencing of noise spot arranged, this method will be subjected to certain influence.Just because of when pretreatment, removed noise, so on the periodicity of estimating gait, can use this method.We find: convex hull area minimum occurs at the 11st, 23,35 and 48 frames, convex hull area maximum occurs at the 6th, 17,29 and 42 frames.This method and method 1 have similar conclusion so, and different with method 1 is: what method 1 was calculated is the pixel number in actual human body zone, and this method be similar to human region, is the number of asking the interior pixel of convex hull.Therefore, this method (as Fig. 5 (2)) ratio method 1 (as Fig. 5 (1)) all has bigger area for each frame, but periodically constant substantially.

2.1.3 observe the periodicity of gait according to the cyclically-varying of the area behind the movement human zone filling cavity in every two field picture

This method can solve the cavity blemish problem that has the gait profile in the image pretreatment in early stage, but is that the situation of real gait profile is also filled its cavity really for this cavity that occurs in the 10th, 12,22 and 49 frames that is similar to simultaneously.4 frames among Fig. 6 are respectively the 10th among Fig. 3,12,22 and 49 frames from left to right, and circle has marked the cavity that occurs on the true gait profile.The design sketch of the method determination cycles such as Fig. 5 (3), the design sketch 5 (1) of it and method 1 is very close, also presents obvious periodic.

Change the periodicity of observing gait 2.1.4 account for the ratio of convex hull according to movement human in every two field picture

This method is the comprehensive of method 2.1.1 and method 2.1.2.Shown in Fig. 5 (4), in the 6th, 17,31 and 42 frames, local minimum has appearred in this ratio; And in the 12nd, 24,36 and 48 frames, local maximum has appearred in this ratio.In like manner can judge from the 6th frame to the 30 frames, from the 17th frame to the 41 frames, from the 12nd frame to the 35 frames, all be independent gait cycle from the 24th frame to the 47 frames.

2.2 the gait cycle detecting method of----barycenter based on provincial characteristics

The horizontal stroke of region R barycenter, vertical coordinate are:

$\stackrel{\‾}{r}=\frac{1}{A}\underset{(r,c)\∈R}{\mathrm{\Σ}}r---\left(3\right)$

$\stackrel{\‾}{c}=\frac{1}{A}\underset{(r,c)\∈R}{\mathrm{\Σ}}c---\left(4\right)$

Shown in Fig. 5 (5), (6), be respectively barycenter vertical coordinate and abscissa situation of change frame by frame, the abscissa by barycenter changes to determine that the cycle is obvious not as good as the vertical coordinate effect with barycenter.So here the cycle of gait is judged in our employing according to the situation of the vertical coordinate variation of barycenter, we find to have occurred at the 11st, 23,35,48 frames the local minizing point of barycenter vertical coordinate, so the 11st frame to the 34 frames are one-periods, the 23rd frame to the 47 frames also are one-periods.But, this method with come the method for determination cycles that some discrepancy are arranged according to the intrinsic area of human body, but little.And we detect by an unaided eye, and the shape difference of movement human also is little in the 47th frame and the 48th frame.

2.3 the gait cycle detecting method of----fitted ellipse based on provincial characteristics (or circle)

2.3.1 according to having the long and short axle of the ellipse of identical second order spatial moment, the periodicity that eccentricity is observed gait with the graphics field

The second order spatial moment in zone has three, is expressed as the capable square μ of second order respectively _Rr, second order row square μ _CcWith second order mixed moment μ _Rc, be defined as follows:

${\mathrm{\μ}}_{\mathrm{rr}}=\frac{1}{A}\underset{(r,c)\∈R}{\mathrm{\Σ}}{(r-\stackrel{\‾}{r})}^2---\left(5\right)$

${\mathrm{\μ}}_{\mathrm{cc}}=\frac{1}{A}\underset{(r,c)\∈R}{\mathrm{\Σ}}{(c-\stackrel{\‾}{c})}^2---\left(6\right)$

${\mathrm{\μ}}_{\mathrm{rc}}=\frac{1}{A}\underset{(r,c)\∈R}{\mathrm{\Σ}}(r-\stackrel{\‾}{r})(c-\stackrel{\‾}{c})---\left(7\right)$

μ _RrExpression departs from the capable variation of capable average, μ _CcExpression departs from the column variation of column mean, μ _RcRepresent off-centered ranks variation, they do not change with the translation and the dimensional variation of two-dimensional shapes, therefore are usually used in describing simple shape.

If region R is oval, it is centered close to initial point, and then R can be expressed as:

R＝{(r，c)|dr ²+2erc+fc ²≤1} (8)

The then coefficient d of elliptic equation, e and f and second moment μ _Rr, μ _CcAnd μ _RcBetween the pass be:

$\left(\begin{array}{cc}d& e\\ e& f\end{array}\right)=\frac{1}{4({\mathrm{\μ}}_{\mathrm{rr}}{\mathrm{\μ}}_{\mathrm{cc}}-{\mathrm{\μ}}_{\mathrm{rc}}^2)}\left(\begin{array}{cc}{\mathrm{\μ}}_{\mathrm{cc}}& -{\mathrm{\μ}}_{\mathrm{rc}}\\ -{\mathrm{\μ}}_{\mathrm{rc}}& {\mathrm{\μ}}_{\mathrm{rr}}\end{array}\right)---\left(9\right)$

Coefficient d, e and f that elliptic equation has been arranged, we can determine oval long and short axle and direction thereof, because elliptic equation coefficient and second moment μ _Rr, μ _CcAnd μ _RcHave above-mentioned relation, so we are by μ _Rr, μ _CcAnd μ _RcCan determine oval long and short axle and direction thereof, discuss as shown in table 1ly that note: following deflection is the direction that rotates counterclockwise from longitudinal axis edge.

Table 1 calculates the long and short axle and the direction thereof of fitted ellipse according to the second order spatial moment

Fig. 7 annotates the location of oval location and long and short axle, and oval deflection is the dotted line of level and the angle of transverse.According to having the long and short axle of ellipse of identical second order spatial moment and the periodic law that eccentricity is sought gait with the graphics field, experimental result is respectively shown in Fig. 5 (7), (8) and (9).Because when extremity close up, the minor axis of fitted ellipse is the shortest, the eccentricity maximum, and the major axis of fitted ellipse be expert at People's Bank of China walk in the process to change not obvious, so Fig. 5 (8) and to scheme the effect of (9) relatively good has periodicity clearly.In Fig. 5 (8), the local minimum situation of minor axis appears in the 11st, 23,35 and 47 frames, also is that the local maximum situation appears in eccentricity just simultaneously.We find that the length of the 47th frame and the 48th frame minor axis is very approaching, and why this difference can occur if just having annotated well based on the method for barycenter with based on intrinsic Method for Area when judging gait cycle.

2.3.2 judge the periodicity of gait according to the situation of change of match diameter of a circle

We go to the zone of match movement human with a circle, make area of this circle equate with the area of human region, use the periodicity of observing gait with respect to the diameter of a circle of every frame human region so, our Practical Calculation be

Wherein: Area represents the area of human region, also is the pixel number that this zone is contained.Be depicted as the design sketch of observing gait cycle by the match diameter of a circle as Fig. 5 (10), it and Fig. 5 (1) shape are similar, and this method is consistent with the cycle location of judging based on intrinsic area.

2.4 the gait cycle detecting method of----some particular points based on provincial characteristics

Fig. 8 defines eight particular point: top-left, top-right, left-top, right-top, left-bottom, right-bottom, bottom-left and bottom-right, and wherein (2) are the special circumstances of (1).

Because in the process that the people walks, extremity rock back and forth with respect to mass center of human body, judge that the cyclically-varying of gait is effective so adopt right limb border, be respectively shown in Fig. 5 (11), (12) that the right-top point is horizontal, vertical coordinate is with the situation of change of gait sequence frame, vertical coordinate with this point comes the sense cycle effect bad as can be seen, so adopting the situation of change of the abscissa of this point judges, the 12nd frame to the 35 frames and the 24th frame to the 47 frames all are respectively independent gait cycles, conclusion and method 1 basically identical that this cycle is judged; Be respectively shown in Fig. 5 (13), (14) that the right-bottom point is horizontal, vertical coordinate is with the situation of change of gait sequence frame, we still come the sense cycle effect bad with the vertical coordinate of this point as can be seen.The position on left-bottom and all corresponding left limb of left-top point major part border, therefore by these 2 cyclically-varyings that also can find out gait, be respectively left-bottom abscissa, left-bottom vertical coordinate, left-top abscissa and left-top vertical coordinate situation of change as Fig. 5 (15), (16), (17), (18) with the gait sequence frame, though the cycle effect is bad, put the cyclically-varying of gait as can be seen by these.Judge the cyclically-varying of gait according to the change in location situation of some particular points in every two field picture movement human zone, though this method is simple, but must guarantee that the pretreatment in early stage is enough good, if having a large amount of noise spots in the image, will cause the segmentation errors of gait cycle.

2.5 the gait cycle detecting method of----bounding box based on provincial characteristics

Sometimes where be positioned at piece image in order to understand a zone roughly, at this moment to use the bounding box in zone, bounding box is a rectangle, by level and vertically four edges whole zone is surrounded, and join with the going up most of zone, the most following, the most left and the rightest point.Be illustrated in figure 9 as the bounding box of human region.

The periodically variable situation of observing gait with the wide variety of bounding box in every frame is shown in Fig. 5 (19), because its local minimum point is very little with near the difference of its point, so we estimate the periodicity of gait with its local maximum point (for example: corresponding to the 6th, 17,29 and 43 frames), all are independent gait cycles from the 6th frame to the 28 frames with from the 17th frame to the 42 frames.We can also judge the periodicity of gait according to the variation that the movement human zone accounts for the bounding box ratio, shown in Fig. 5 (20).Near local minimum in the extreme value and their some difference is less, so use local minimum to put to judge and improper.And near the point the local maximum point in the extreme value and their differs greatly, and when the 12nd, 24,35 and 48 frames, the local maximum of this ratio occurred, so all be independent gait cycle from the 12nd frame to the 34 frames with from the 24th frame to the 47 frames.

The various gait cycle detecting methods of 3 contrasts

Sum up the above simple and effective gait cycle detecting method of five classes, wherein feasible 10 kinds are summarized as table 2 based on regional characteristics analysis (comprising: intrinsic area, convex hull area, go empty area, human body to account for convex hull ratio, barycenter vertical coordinate, fitted ellipse minor axis, fitted ellipse eccentricity, match diameter of a circle, right-top (bottom) abscissa and human body to account for bounding box ratio or the like) determination cycles characteristic.Their advantage is to calculate easy, and amount of calculation is far smaller than document [1] [2] [3] [4], realizes easily, and has reached the precision of people's subjective judgment.The gait cycle signal that convex hull area-method, fitted ellipse minor axis method, fitted ellipse eccentricity method, right-top (bottom) abscissa method obtain is all smoother.Intrinsic area-method, to go empty area-method, barycenter vertical coordinate method, fitted ellipse minor axis method, fitted ellipse eccentricity method, match circular diameter method and human body to account for the bounding box rule of three stronger to the robustness of noise, and convex hull area-method, right-top (bottom) abscissa method and human body account for the convex hull rule of three to the robustness of noise a little less than, so when these three kinds of methods of employing are carried out cycle detection, must carry out the pretreatment work of image sequence earlier.Remove except the minor axis and eccentricity method of fitted ellipse, other method all must be carried out after pretreated center for standardization, and because the minor axis and the eccentricity method of fitted ellipse have yardstick, translation invariance, the method for these two kinds of cycle detections can be carried out before pretreated center for standardization.Intrinsic area-method, the cycle of going empty area-method, fitted ellipse minor axis method, fitted ellipse eccentricity method, match diameter of a circle method and right-top (bottom) abscissa method all can obtain this gait sequence are the conclusions of 24 frames, and there is the conclusion of judging 23 or 25 frames in additive method.These methods judge that it also is normal that difference appears in gait cycle, even because perusal also is difficult to distinguish the difference of certain two interframe.

Table 2 the whole bag of tricks is judged the summary of gait cycle characteristic

This patent is converted into the graphics field feature analysis problem of single frames with the gait cycle problem analysis, has proposed five classes based on the regional characteristics analysis gait cycle detecting method of (comprising: the ellipse of area, barycenter, match (or circle), some particular points and bounding box).Comprising 10 kinds of simple and feasible gait detection methods: intrinsic area-method, convex hull area-method, go empty area-method, human body to account for convex hull rule of three, barycenter vertical coordinate method, fitted ellipse minor axis method, fitted ellipse eccentricity method, match diameter of a circle method, right-top (bottom) abscissa method and human body to account for bounding box rule of three or the like.They have all reached the precision of artificial judgement gait cycle.Performance the best of fitted ellipse minor axis method and fitted ellipse eccentricity method, the gait cycle signal smoothing that not only obtains, stronger to the robustness of noise, and have yardstick, a translation invariance, the method of these two kinds of cycle detections can be carried out before pretreated center for standardization, reduced the amount of calculation of other periodic frames so widely, and shortened the time that Gait Recognition is dealt with the work in earlier stage about Flame Image Process.

Claims (1)

1. gait cycle detecting method based on regional characteristics analysis comprises that obtaining with gait cycle of pedestrian's objective contour detect; It is characterized in that: the method for obtaining of described pedestrian's objective contour is, at first from video, extract single-frame images and carry out greyscale transformation, calculate the intermediate value of each pixel in frame by frame then, background image as whole sequence, adopt the background subtraction method to extract human body target at last, fill up cavity, the simply connected analysis of binary image with mathematical morphology and extract people's silhouette, make human body placed in the middle, the size of image is unified to be the 64*64 pixel; It is the graphics field feature analysis problem that the gait cycle problem analysis is converted into single frames that described gait cycle detects, promptly analyze the cycle of gait, be a gait cycle to occurring local extremum for the third time again from local extremum occurring for the first time according to the changing features situation of graphics field in every frame; Described graphics field feature is the fitted ellipse of graphics field in every frame; Described fitted ellipse is the long and short axle and the eccentricity of the ellipse of human region with identical second order spatial moment, according to having the long and short axle of the ellipse of identical second order spatial moment, the periodicity that eccentricity is observed gait with the graphics field,

The second order spatial moment in zone has three, is expressed as the capable square μ of second order respectively _Rr, second order row square μ _CcWith second order mixed moment μ _Rc, be defined as follows:

${\mathrm{\μ}}_{\mathrm{rr}}=\frac{1}{A}\underset{(r,c)\∈R}{\mathrm{\Σ}}{(r-\stackrel{\‾}{r})}^2$

${\mathrm{\μ}}_{\mathrm{cc}}=\frac{1}{A}\underset{(r,c)\∈R}{\mathrm{\Σ}}{(c-\stackrel{\‾}{c})}^2$

${\mathrm{\μ}}_{\mathrm{rc}}=\frac{1}{A}\underset{(r,c)\∈R}{\mathrm{\Σ}}(r-\stackrel{\‾}{r})(c-\stackrel{\‾}{c})$

A represents the intrinsic area in zone, Be the abscissa of region R barycenter,

Vertical coordinate for the region R barycenter; μ _RrExpression departs from the capable variation of capable average, μ _CcExpression departs from the column variation of column mean, μ _RcRepresent off-centered ranks variation, they do not change with the translation and the dimensional variation of two-dimensional shapes,

If region R is oval, it is centered close to initial point, and then R is expressed as:

R＝{(r，c)|dr ²+2erc+fc ²≤1}

The then coefficient d of elliptic equation, e and f and second moment μ _Rr, μ _CcAnd μ _RcBetween the pass be:

$\left(\begin{array}{cc}d& e\\ e& f\end{array}\right)=\frac{1}{4({\mathrm{\μ}}_{\mathrm{rr}}{\mathrm{\μ}}_{\mathrm{cc}}-{\mathrm{\μ}}_{\mathrm{rc}}^2)}\left(\begin{array}{cc}{\mathrm{\μ}}_{\mathrm{cc}}& {-\mathrm{\μ}}_{\mathrm{rc}}\\ {-\mathrm{\μ}}_{\mathrm{rc}}& {\mathrm{\μ}}_{\mathrm{rr}}\end{array}\right)$

Coefficient d, e and f that elliptic equation has been arranged determine oval long and short axle and direction thereof, because elliptic equation coefficient and second moment μ _Rr, μ _CcAnd μ _RcHas above-mentioned relation, therefore by μ _Rr, μ _CcAnd μ _RcDetermine oval long and short axle and direction thereof.

Images