CN103886605A - Method for predicting and tracking moving object based on center of curvature - Google Patents

Abstract

The invention discloses a method for predicting and tracking a moving object based on the center of curvature. The method comprises the steps that a moving object in video is processed and the center of curvature of the moving target in a frame of image is calculated through a support function; a time axis is added to the center of curvature and the center of curvature is expanded to a time-space domain; a motion prediction model is established and the specific position of the center of curvature of a next frame of the moving object can be predicted through the current frame; a mapping model is established and an object center-of-curvature coordinate obtained through two-dimensional prediction is converted into three-dimensional real space. The effects that the moving object is predicted and tracked are achieved through the method based on the center of curvature, a movement track of the object can be rapidly and accurately predicted, and the real movement track of the object in the three-dimensional space can be reflected.

Description

A kind of prediction based on the center of curvature and the method for following the tracks of mobile object

[technical field]

The present invention relates to video tracking field, particularly a kind of prediction based on the center of curvature (Steiner point) and the method for tracked mobile target.

[background technology]

The detection and tracking of the moving target based on unique point are the important contents of computer vision research, and it all has a wide range of applications in fields such as Aero-Space, intelligent robot, automatic monitored control system, medical image analysis and video compress.Unique point can represent part or the global information of moving object, can well react the operation information of object.But, at present, these existing track algorithms based on unique point have two deficiencies: 1) these target tracking algorisms based on unique point can only trace into the movement locus of 2D plane, and often two dimensional image plane can not reflect the movement locus of object in real space.2) in target following, existing method is calculated the unique point of each frame and is followed the tracks of, and priori does not use like this, has increased the complicacy of calculating.

The geometric center of an object of Steiner point thing.Defined in nineteen sixty by Shephard, hereafter, Steiner point start to be concerned and with it solve various problems.It is only relevant with the convex closure of object, and can be by supporting that function calculates.And the computational complexity of support function is O (n).Steiner point has good algebraic property, such as continuity, rotational invariance, additive property etc.

Realizing in process of the present invention, inventor utilizes Steiner point to predict and tracked mobile target, mainly solves two problems above-mentioned, that is: 1) find the mapping function of 2 d-to-3 d, can rediscover space in the track of moving target.2) reduce the complexity of calculating, utilize priori to predict the position of object, improve the accuracy of prediction.

[summary of the invention]

The object of the present invention is to provide a kind of prediction based on the center of curvature (Steiner point) and the method for tracked mobile target, the described method of moving target being predicted and being followed the tracks of based on Steiner point, the motion track of object can be predicted fast and accurately, and object real movement locus in three dimensions can be reacted by mapping function.

In order to reach object of the present invention, according to an aspect of the present invention, the invention provides a kind of method of prediction of ordering based on Steiner and tracked mobile target, described method comprises: moving target in video is processed, calculated the Steiner point of moving target in a two field picture by support function; Steiner point based on obtaining adds time shaft, is generalized on time-space domain; Set up motion prediction model, the particular location that can order by the Steiner of predicted current frame next frame moving target; Set up mapping model, two-dimensional prediction to target Steiner point coordinate be mapped in three-dimensional realistic space.

For a two field picture, P is the convex closure of a mobile object in an image; It is P={p that convex closure P has M summit ₁, p ₂... p _m; By support function can the Steiner point of object P be:

Wherein p _jsummit P _jcoordinate position,

p _jexterior angle value, and meet

Further, the Steiner point obtaining based on upper step adds time shaft, is generalized on time-space domain, and the mobile Steiner point on space-time spatial domain is:

$S(A:t)=\frac{1}{V\left(B^n\right)}{\∫}_{S^{n-1}}{h}_A(e,t)\mathrm{ed\λ}\left(e\right)$

Wherein A is object mobile in video; S ^n-1it is unit ball; E is the vector of unit length on unit ball; λ is based on S ^n-1on legesgue estimate; V (B ⁿ) be unit ball B ⁿon volume; T is time shaft.

Further, set up motion prediction model, the particular location that can order by the Steiner of predicted current frame next frame moving target.If position vector x _kat time t _kin time, is made up of five dimensional vectors, is respectively Steiner point coordinate position (p _{x, k}, p _{y, k}), change angle ψ _k, the movement velocity v of object _kwith moving radius R _k, that is:

We can suppose that object of which movement is based on circular arc type, so can pass through t _kposition is that the coordinate that the Steiner of object is ordered predicts that next frame is t _k+1the position that when moment, the Steiner of object is ordered:

${\mathrm{\υ}}_{k+1}=\frac{\sqrt{{(p_{x,k+1}-p_{x,k})}^2+{(p_{y,k+1}-p_{y,k})}^2}}{\mathrm{\τ}};$

$R_k=\frac{{(1+{{\mathrm{\υ}}_k}^{\′2})}^{\frac{3}{2}}}{{\mathrm{\υ}}_k^{\′\′}};$

Further, by mapping function, select suitable coordinate system, the coordinate that the two-dimentional Steiner that prediction obtains above can be ordered is mapped in three dimensions:

$\left(\begin{array}{c}{P}_{X_i}\\ P_{X_i}\\ 1\end{array}\right)=H\left(\begin{array}{c}{p}_{x_i}\\ p_{y_i}\\ 1\end{array}\right)$

Wherein (p _xi, p _yi) be the coordinate in two-dimensional space, (P _xi, P _yi, 1) and be the coordinate points in three dimensions.

Mapping matrix H solves: choose at random 4 points, measure four point coordinate values in reality and with its in image for 4 coordinate figures in image, utilize method that singular value solves to solve three minimal eigenvalues of mapping matrix H; And the corresponding proper vector of minimal eigenvalue is require to try to achieve mapping matrix.

[brief description of the drawings]

In conjunction with reference to accompanying drawing and ensuing detailed description, the present invention will be easier to understand, and Fig. 1 is prediction and the tracked mobile target method method flow diagram in one embodiment based on the center of curvature (Steiner point) in the present invention.

[embodiment]

For above-mentioned purpose of the present invention, feature and advantage can be become apparent more, below in conjunction with the drawings and specific embodiments, the present invention is further detailed explanation.

Embodiment of the present invention provides a kind of method of prediction of ordering based on Steiner and tracked mobile target, the method of described prediction of ordering based on Steiner and tracked mobile target, the motion track of object can be predicted fast and accurately, and object real movement locus in three dimensions can be reacted.

Please refer to Fig. 1, its method that shows the prediction of ordering based on Steiner in the present invention and tracked mobile target method flow diagram in one embodiment.Described Eigenvalue Extraction Method 100 comprises:

Step 102, processes and finds convex closure the image of input, utilizes support function to obtain the Steiner point of convex closure;

Use the prediction of ordering based on Steiner and the method for tracked mobile target to testing video used, can predict fast and accurately the motion track of object, and can react object real movement locus in three dimensions.The image of image capture device collection normally divides frame to one section of video, obtain the consecutive image of certain resolution, such as gray level images common 24 frames per second, that resolution is 1024*768 (if not gray level image can be converted into gray level image).In algorithm experimental or practical application, choose a vehicle of turning, choose 137 frames wherein and for continuous.On each frame, the area coverage of vehicle is approximately 23 × 30 pixels.Only need process first three frame, on each frame, the Steiner point of moving vehicle is:

Wherein p _jsummit P _jcoordinate position,

p _jexterior angle value, and meet

Step 104, adds space-time countershaft Steiner point, is generalized on time-space domain;

102 steps are tried to achieve to Steiner point, add time shaft, be generalized to time-space domain and get on, a moving target only has a Steiner point, moves, so add that time domain can reflect mobile Steiner point, the time interval of two two field pictures when object

Step 106, sets up motion prediction model, prediction Steiner point;

Set up motion prediction model by image of the present invention, prediction Steiner point, vehicle is in turning in crossing, and running orbit is approximate is a circular arc.In 102, tried to achieve the position of first three frame Steiner, the speed of vehicle operating is that on this circular arc, the tangential direction of 3 was 10 pixel/seconds.The radius that object is turned is 12.8 pixels.The 3rd frame Steiner point is (3.23,45.57), can obtain the 4th frame Steiner point for (3.14,45.87) by prediction.

Step 108, sets up mapping model, and the two-dimentional Steiner point of prediction is mapped in three dimensions;

Set up mapping model by image of the present invention, on real road, measure four point coordinate and be respectively: (0 0), (0.85 0), (0.85 0.25), (0 0.25).The pixel coordinate in image of its correspondence is: (30.8505 222.3835), (143.8072 214.4660), (143.2794 203.9093), (30.3227 211.2990).By utilizing SVD method to solve three minimal eigenvalues of mapping matrix H; And the corresponding proper vector of minimal eigenvalue is require to try to achieve mapping matrix H.H is:

$H=\left(\begin{array}{ccc}-0.0014& 0.0001& 0.0281\\ 0.0003& 0.0044& -0.9804\\ 0.0001& 0& -0.1950\end{array}\right)$

Therefore can the point of prediction be mapped in three dimensions and be got by mapping matrix.

Further point out, the degree of accuracy of the position prediction of the present invention to vehicle is very high, and while being that physical location compares, error only has 0.2.The degree of accuracy of prediction is higher and the present invention is in turn inside diameter camber during more more close to circular arc! The present invention is by means of the convex closure of object, no matter the distribution of interior of articles pixel, so the complexity of algorithm is very low, is linear! Travelling speed is very fast, is well suited for real-time follow-up.

The prediction of ordering based on Steiner that the present invention proposes and the method for tracked mobile target, can predict the motion track of object fast and accurately, and can react object real movement locus in three dimensions.Classical track algorithm for present: KLT and mean shift.The present invention and they contrast.

By above-mentioned instantiation, the present invention is based on prediction that Steiner orders and the method for tracked mobile target, the motion track of object can be predicted fast and accurately, and object real movement locus in three dimensions can be reacted.

It should be noted that: the prediction of ordering based on Steiner that above-described embodiment provides and the method for tracked mobile target, only be illustrated with the division of above-mentioned each functional module, in practical application, can above-mentioned functions be distributed and completed by different functional modules as required, be four mapping points that input picture, the difference of different resolution chosen, the object of different motion, to complete all or part of function described above.

Above-mentioned explanation has fully disclosed the specific embodiment of the present invention.It is pointed out that and be familiar with the scope that any change that person skilled in art does the specific embodiment of the present invention does not all depart from claims of the present invention.Correspondingly, the scope of claim of the present invention is also not limited only to described embodiment.

Claims (5)

1. the prediction based on the center of curvature (Steiner point) and a method for tracked mobile target, is characterized in that, described method comprises:

Moving target in video is processed, calculated the Steiner point of moving target in a two field picture by support function;

Steiner point based on obtaining adds time shaft, is generalized on time-space domain;

Set up motion prediction model, the particular location that can order by the Steiner of predicted current frame next frame moving target;

Set up mapping model, two-dimensional prediction to target Steiner point coordinate be mapped in three-dimensional realistic space.

2. a kind of method of ordering based on Steiner according to claim 1, is characterized in that, described moving target in video is processed, and calculates the Steiner point of moving target in a two field picture by support function:

For a two field picture, P is the convex closure of a mobile object in an image; It is P={p that convex closure P has M summit ₁, p ₂... p _m; By support function can the Steiner point of object P be:

Wherein p _jsummit P _jcoordinate position,

p _jexterior angle value, and meet

3. the method for a kind of prediction of ordering based on Steiner according to claim 1 and tracked mobile target, is characterized in that, the Steiner point based on obtaining adds time shaft, is generalized on time-space domain, and the mobile Steiner point on time-space domain is

$S(A:t)=\frac{1}{V\left(B^n\right)}{\∫}_{S^{n-1}}{h}_A(e,t)\mathrm{ed\λ}\left(e\right)$

Wherein A is object mobile in video; S ^n-1it is unit ball; E is the vector of unit length on unit ball; λ is S ^n-1on legesgue estimate; V (B ⁿ) be unit ball B ⁿon volume.

4. the method for a kind of prediction of ordering based on Steiner according to claim 1 and tracked mobile target, is characterized in that, sets up motion prediction model, the particular location that can order by the Steiner of predicted current frame next frame moving target:

Position vector x _kat time t _kin time, is made up of five dimensional vectors, is respectively Steiner point coordinate position (p _{x, k}, p _{y, k}), change angle ψ _k, the movement velocity v of object _kwith moving radius R _k, that is:

We can suppose that object of which movement is based on circular arc type, so can pass through t _kposition is that the coordinate that the Steiner of object is ordered predicts that next frame is t _k+1the position that when moment, the Steiner of object is ordered:

${\mathrm{\υ}}_{k+1}=\frac{\sqrt{{(p_{x,k+1}-p_{x,k})}^2+{(p_{y,k+1}-p_{y,k})}^2}}{\mathrm{\τ}};$

$R_k=\frac{{(1+{{\mathrm{\υ}}_k}^{\′2})}^{\frac{3}{2}}}{{\mathrm{\υ}}_k^{\′\′}};$

, by mapping function, select suitable coordinate system, the coordinate that the two-dimentional Steiner that prediction obtains above can be ordered is mapped in three dimensions:

$\left(\begin{array}{c}{P}_{X_i}\\ P_{X_i}\\ 1\end{array}\right)=H\left(\begin{array}{c}{p}_{x_i}\\ p_{y_i}\\ 1\end{array}\right)$

Wherein (p _xi, p _yi) be the coordinate in two-dimensional space, (P _xi, P _yi, 1) and be the coordinate points in three dimensions.

5. mapping function method according to claim 4, is characterized in that, solves mapping matrix H and comprises:

Utilize the method for svd to solve three minimal eigenvalues of mapping matrix H; And the corresponding proper vector of minimal eigenvalue is require to try to achieve mapping matrix.

Images