CN102314694A - Photoelectric navigation motion vector prediction method - Google Patents

Abstract

The invention relates to a photoelectric navigation motion vector prediction method. The photoelectric navigation motion vector prediction method includes the following steps: (a) acquiring two neighboring frames of navigation images, which are respectively formed into a reference frame and a target frame; (b) setting a reference block (B) in the reference frame; (c) setting optional matched modules with the starting point of the reference block B and eight points around the reference block (B) as a center as starting points in the target frame; (d) calculating the sum of the absolute values of the differences between the corresponding pixels in the pixel matrixes of the optional matched modules and the corresponding pixels in the pixel matrix of the reference block (B), so that the optimal matched module can be obtained when the sum of the absolute value of the differences is minimum; (e) working out a corrected motion vector value (Alpha); (f) selecting three motion vectors (Vn-3, Vn-2 and Vn-1) before the target frame, and according to a target function, working out a vector value on the X axis; (g) according to a target function, working out a vector value on the Y axis; (h) finishing the prediction of the motion vector of the target frame. The method is easy to operate, has high prediction precision, and is highly controllable, safe and reliable.

Description

Photoconduction shipping dynamic vector Forecasting Methodology

Technical field

The present invention relates to a kind of vector prediction method, especially a kind of photoconductive shipping dynamic vector Forecasting Methodology belongs to the technical field that photoelectricity navigates.

Background technology

Formula V is used in the estimation of the motion vector of photoelectricity navigational system _n=pre_v+a representes, wherein V _nBe the output valve of current motion vector, pre_v is the predicted value of current motion vector, equals the displacement vector of last computing gained, i.e. pre_v=V _N-1A is in order to regulating the modified value of current motion vector, and a is the relative displacement of the module calculated through the mode of images match.The shortcoming of prior art is as last motion vector V _N-1During miscount, current motion vector output valve also is wrong, produces chain reaction like this and so wrong constantly amplification is finally caused the uncontrollable of navigational system.

Summary of the invention

The objective of the invention is to overcome the deficiency that exists in the prior art, a kind of photoconductive shipping dynamic vector Forecasting Methodology is provided, its method is simple to operate, and precision of prediction is high, and controllability is good, and is safe and reliable.

According to technical scheme provided by the invention, a kind of photoconductive shipping dynamic vector Forecasting Methodology, said method of motion vector prediction comprises the steps:

A, the adjacent two frame navigation pictures of high speed acquisition, the former frame in the said two frame navigation pictures forms reference frame, and another two field picture forms target frame;

B, reference block B is set in reference frame, the picture element matrix of reference block B is F _B, said picture element matrix F _BSize be m * n, the origin coordinates of reference block B is (x ₀, y ₀);

C, in target frame, be provided with (x ₁, y ₁) and with its be the center around 8 points be starting point, size be 9 modules of m * n as the match selection module, the picture element matrix of said match selection module is respectively F ₁～F ₉

D, according to the picture element matrix of 9 match selection modules that obtain among the step c, calculate the picture element matrix F of match selection module respectively ₁～F ₉With F in the picture element matrix of reference block B _BBetween respective pixel difference absolute value with, promptly

${\mathrm{Sum}}_{\mathrm{Bi}}=\underset{x=0}{\overset{x=m}{\mathrm{\Σ}}}\underset{y=0}{\overset{y=n}{\mathrm{\Σ}}}|F_B(x_0+x,y_0+y)-F_i(x_i+x,y_i+y)|,i=\mathrm{1,2}\mathrm{\Λ},9;$

And as said Sum _BiHour, it is the optimum matching module that corresponding matched is selected module;

E, according to the relative position between optimum matching module that obtains in the steps d and reference block B, obtain motion vector modified value α;

F, choose three motion vector V before the target frame _N-3, V _N-2, V _N-1, the motion vector of target frame is pre_v; The maximal value of setting in above-mentioned three motion vectors of x direction of principal axis is p_x=max ([V _{N-3_x}, V _{N-2_x}, V _{N-1_x}]), the minimum value on the x direction of principal axis in above-mentioned three motion vectors is q_x=min ([V _{N-3_x}, V _{N-2_x}, V _{N-1_x}]); According to objective function S _Min=| pre_vx-V _{N-3_x}|+| pre_vx-V _{N-2_x}|+| pre_vx-V _{N-1_x}|, when said objective function is got minimum value, obtain the vector value pre_vx of target frame on the x axle, pre_vx ∈ [q_x, p_x];

G, choose three motion vector V before the target frame _N-3, V _N-2, V _N-1, the motion vector of target frame is pre_v; The maximal value of setting above-mentioned three motion vectors of y direction of principal axis is p_y=max ([V _{N-3_y}, V _{N-2_y}, V _{N-1_y}]), the minimum value on the y direction of principal axis in above-mentioned three motion vectors is q_y=min ([V _{N-3_y}, v _{N-2_y}, V _{N-1_y}]); According to objective function S _Min=| pre_vy-V _{N-3_y}|+| pre_vy-V _{N-2_y}|+| pre_vy-V _{N-1_y}|, when said objective function is got minimum value, obtain the vector value pre_vy of target frame on the y axle, pre_vy ∈ [q_y, p_y];

H, obtain the motion vector pre_v of target frame, and, accomplish the prediction of target frame motion vector according to the motion vector modified value α that step e obtains according to step f and step g.

In said step f and the step g, work as V _N-3=V _N-1The time, pre_v=V _N-1In said step f and the step g, work as V _N-3≠ V _N-1, and work as V _N-3=V _N-2The time, pre_v=V _N-2In said step f and the step g, work as V _N-3≠ V _N-1, and work as V _N-3≠ V _N-2The time, pre_v=V _N-2

Advantage of the present invention:, can access reference frame and target frame through the high speed acquisition navigation picture; Through selection reference piece B in reference frame, in target frame, select the match selection module, select to select module with the optimum matching of reference block B, can access the modified value of target frame motion-vector prediction; Then according to the motion vector of three frame motion vector value prediction target frame before the target frame, the effectively estimation of controlled target frame motion vector when being out of one's reckoning, can being regulated effectively and not make mistake transmission; Method is simple to operate, and precision of prediction is high, and controllability is good, and is safe and reliable.

Description of drawings

Fig. 1 chooses the synoptic diagram of choosing reference block in match selection module and the reference frame in the target frame of the present invention.

Fig. 2 is a kind of process flow diagram of the present invention to the target frame motion-vector prediction.

Fig. 3 is the another kind of process flow diagram of the present invention to the target frame motion-vector prediction.

Embodiment

Below in conjunction with concrete accompanying drawing and embodiment the present invention is described further.

In order to improve the precision of motion-vector prediction in the photoelectricity navigation, guarantee the reliability of navigational system, the present invention's photoconduction shipping dynamic vector Forecasting Methodology comprises the steps:

A, the adjacent two frame navigation pictures of high speed acquisition, the former frame in the said two frame navigation pictures forms reference frame, and another two field picture forms target frame;

During said high speed acquisition two frame navigation pictures, the side-play amount of adjacent two two field pictures is not too large, and the modified value of movement velocity is less;

B, reference block B is set in reference frame, the picture element matrix of reference block B is F _B, said picture element matrix F _BSize be m * n, the origin coordinates of reference block B is (x ₀, y ₀);

C, in target frame, be provided with (x ₁, y ₁) and with its be the center around 8 points be starting point, size be 9 modules of m * n as the match selection module, the picture element matrix of said match selection module is respectively F ₁～F ₉

As shown in Figure 1: as to select module F in target frame, selecting corresponding matched ₁～F ₉Behind the high speed acquisition navigation picture, match selection module F ₁～F ₉And have correspondence and matching between reference block B, select module can access the motion vector modified value of target frame through selecting optimum matching; In target frame, select match selection module F ₁In, because target frame and reference frame overlapping and choose relation can obtain origin coordinates (x ₁, y ₁) with the origin coordinates of reference block B be (x ₀, y ₀) have an identical coordinate figure, match selection module F ₂～F ₉Other origin coordinates in target frame choose with target frame in select match selection module F ₁Corresponding.

D, according to the picture element matrix of 9 match selection modules that obtain among the step c, calculate the picture element matrix F of match selection module respectively ₁～F ₉With F in the picture element matrix of reference block B _BBetween respective pixel difference absolute value with, promptly

${\mathrm{Sum}}_{\mathrm{Bi}}=\underset{x=0}{\overset{x=m}{\mathrm{\Σ}}}\underset{y=0}{\overset{y=n}{\mathrm{\Σ}}}|F_B(x_0+x,y_0+y)-F_i(x_i+x,y_i+y)|,i=\mathrm{1,2}\mathrm{\Λ},9;$

And as said Sum _BiHour, it is the optimum matching module that corresponding matched is selected module; As above-mentioned Sum _BiHour, have maximum matching degree between the match selection module that can confirm to obtain choosing in the target frame and reference block B, thereby can confirm the motion vector modified value α between target frame and the reference frame; The picture element matrix F of match selection module ₁～F ₉Starting point all at the starting point (x of reference block B ₀, y ₀) on the corresponding x axle, y direction of principal axis ± 1 in, can make things convenient for accurately to obtain motion vector modified value α;

E, according to the relative position between optimum matching module that obtains in the steps d and reference block B, obtain motion vector modified value α;

Said motion vector modified value α is included in the modified value on x axle and the y direction of principal axis, according to the relation of choosing between match selection module and reference block B, can obtain motion vector modified value α and be [1,1] at x axle and the axial correction scope of y;

F, choose three motion vector V before the target frame _N-3, V _N-2, V _N-1, the motion vector of target frame is pre_v; The maximal value of setting in above-mentioned three motion vectors of x direction of principal axis is p_x=max ([V _{N-3_x}, V _{N-2_x}, V _{N-1_x}]), the minimum value on the x direction of principal axis in above-mentioned three motion vectors is q_x=min ([V _{N-3_x}, V _{N-2_x}, V _{N-1_x}]); According to objective function S _Min=| pre_vx-V _{N-3_x}|+| pre_vx-V _{N-2_x}|+| pre_vx-V _{N-1_x}|, when said objective function is got minimum value, obtain the vector value pre_vx of target frame on the x axle, pre_vx ∈ [q_x, p_x];

When navigation picture was carried out high-speed sampling, the conversion of the moving displacement vector of continuous multiple frames image was less, can not occur basically quickening continuously or the deceleration situation; When if the variation of predictive vector is excessive, makeover process can't be withdrawn into the deviation range of allowing to motion vector; In order to improve forecasting reliability, need target frame three frame motion vectors before to carry out reference prediction.Said three frame motion vectors of choosing are as much as possible near present frame; When considering the high speed acquisition image; Corresponding data can not produce under the situation of sudden change; Based on choosing the relation of three frame reference frames on the x direction of principal axis; Based on the relation of respective, can access the predictive vector value of above-mentioned target frame at the x axle;

G, choose three motion vector V before the target frame _N-3, V _N-2, V _N-1, the motion vector of target frame is pre_v; The maximal value of setting above-mentioned three motion vectors of y direction of principal axis is p_y=max ([V _{N-3_y}, V _{N-2_y}, V _{N-1_y}]), the minimum value on the y direction of principal axis in above-mentioned three motion vectors is q_y=min ([V _{N-3_y}, V _{N-2_y}, V _{N-1_y}]); According to objective function S _Min=| pre_vy-V _{N-3_y}|+| pre_vy-V _{N-2_y}|+| pre_vy-V _{N-1_y}|, when said objective function is got minimum value, obtain the vector value pre_vy of target frame on the y axle, pre_vy ∈ [q_y, p_y];

As shown in Figure 2, according to step f, in like manner can access the predictive vector value of target frame at the y axle; Pre_vx and pre_vy all adopt the shaping data when prediction;

H, obtain the motion vector pre_v of target frame, and, accomplish the prediction of target frame motion vector according to the motion vector modified value α that step e obtains according to step f and step g; Thereby the predicted value according to above-mentioned target frame motion vector can improve controllability and reliability in the photoelectricity navigational system.

To the prediction of target frame motion vector, can also obtain through following method step.Particularly, as shown in Figure 3.As above-mentioned three motion vector V that choose _N-3, V _N-2, V _N-1Vector value when all unequal, explain that image sensor doing to quicken computing always, and this situation is unallowed in the photoelectricity navigation, we can think thus must have one to be out of one's reckoning in selected three motion vectors.Use wrong amount and make the estimation generation deviation of movement locus for fear of the process of carrying out pre_v prediction, therefore select for use the intermediate value that least possibly make a mistake to predict, concrete forecasting process is: said three motion vector V that choose _N-3, V _N-2, V _N-1In, work as V _N-3=V _N-1The time, pre_v=V _N-1Work as V _N-3≠ V _N-1, and work as V _N-3=V _N-2The time, pre_v=V _N-2Work as V _N-3≠ V _N-1, and work as V _N-3≠ V _N-2The time, pre_v=V _N-2Based on above-mentioned prediction, the prediction that can guarantee target frame can not make navigation path generation deviation.

Embodiment

With a concrete embodiment forecasting process of the present invention is described below.

If V _{N_x}Be the motion vector on the x direction of principal axis, V _{N_y}Motion vector on the y direction of principal axis then has $\begin{array}{c}{V}_{n\_x}=\mathrm{Pre}\_\mathrm{Vx}+\mathrm{\α}\_x\\ V_{n\_y}=\mathrm{Pre}\_\mathrm{Vy}+\mathrm{\α}\_y\end{array}$ Wherein pre_vx is the predicted value of target frame motion vector on the x direction of principal axis; α _ x is the modified value of target frame motion vector on the x direction of principal axis; Pre_vy is the predicted value of target frame motion vector on the y direction of principal axis, and α _ y is the modified value of target frame motion vector on the y direction of principal axis.

According to the method for 9 couplings, the value of estimation α _ x and α _ y when supposing that the matching module best with it is F7, can access then α _ x=1, α _ y=-1.

Prediction to pre_vx and pre_vy

Existing method:

V according to last time calculating _{N-1_x}And V _{N-1_y}Motion vector to present frame is predicted, that is to say that between current adjacent two frames of hypothesis be with V on the x direction of principal axis _{N-1_x}With on the x direction of principal axis with V _{N-1_y}Move with uniform velocity, promptly $\begin{array}{c}\mathrm{Pre}\_\mathrm{Vx}=V_{n-1\_x}\\ \mathrm{Pre}\_\mathrm{Vy}=V_{n-1\_y}\end{array}.$

First kind of Forecasting Methodology of the present invention:

If adjacent three times motion vector is respectively: $\begin{array}{c}{V}_{n-3\_x},V_{n-2\_x},V_{n-1\_x}\\ V_{n-3\_y},V_{n-2\_y},V_{n-1\_y}\end{array},$ Predictive vector is pre_vx and pre_vy.The value of pre_vx is [q_x, p_x] its interior integer data, wherein p_x=max ([V _{N-3_x}, V _{N-2_x}, V _{N-1_x}]), q_x=min ([V _{N-3_x}, V _{N-2_x}, V _{N-1_x}]).In [q_x, p_x] included interval, find out a corresponding round values so that the objective function Smin=|pre_vx-V of said setting _{N-3_x}|+| pre_vx-V _{N-2_x}|+| pre_vx-V _{N-1_x}| minimum, the value of this moment is pre_vx.In like manner can draw the value of pre_vy.

Second kind of Forecasting Methodology of the present invention:

(1), works as V _{N-3_x}=V _{N-1_x}The time, pre_vx=V _{N-1_x}

(2), work as V _{N-3_x}≠ V _{N-1_x}The time,

A. work as V _{N-3_x}=V _{N-2_x}The time, pre_vx=V _{N-2_x}

B. work as V _{N-3_x}≠ V _{N-2_x}The time, pre_vx=V _{N-2_x}

Calculating on the Y direction is the same; Thereby can access the motion vector predictor of target frame.

The present invention can access reference frame and target frame through the high speed acquisition navigation picture; Through selection reference piece B in reference frame, in target frame, select the match selection module, select to select module with the optimum matching of reference block B, can access the modified value of target frame motion-vector prediction; Then according to the motion vector of three frame motion vector value prediction target frame before the target frame, the effectively estimation of controlled target frame motion vector when being out of one's reckoning, can being regulated effectively and not make mistake transmission; Method is simple to operate, and precision of prediction is high, and controllability is good, and is safe and reliable.

Claims (4)

1. a photoconductive shipping dynamic vector Forecasting Methodology is characterized in that said method of motion vector prediction comprises the steps:

(a), the adjacent two frame navigation pictures of high speed acquisition, former frame in the said two frame navigation pictures forms reference frame, another two field picture forms target frame;

(b), reference block B is set in reference frame, the picture element matrix of reference block B is F _B, said picture element matrix F _BSize be m * n, the origin coordinates of reference block B is (x ₀, y ₀);

(c), in target frame, be provided with (x ₁, y ₁) and with its be the center around 8 points be starting point, size be 9 modules of m * n as the match selection module, the picture element matrix of said match selection module is respectively F ₁～F ₉

(d), according to the picture element matrix of 9 match selection modules that obtain in the step (c), calculate the picture element matrix F of match selection module respectively ₁～F ₉With F in the picture element matrix of reference block B _BBetween respective pixel difference absolute value with, promptly

${\mathrm{Sum}}_{\mathrm{Bi}}=\underset{x=0}{\overset{x=m}{\mathrm{\Σ}}}\underset{y=0}{\overset{y=n}{\mathrm{\Σ}}}|F_B(x_0+x,y_0+y)-F_i(x_i+x,y_i+y)|,i=\mathrm{1,2}\mathrm{\Λ},9;$

And as said Sum _BiHour, it is the optimum matching module that corresponding matched is selected module;

(e), according to optimum matching module that obtains in the step (d) and the relative position between reference block B, obtain motion vector modified value α;

(f), choose target frame three motion vector V before _N-3, V _N-2, V _N-1, the motion vector of target frame is pre_v; The maximal value of setting in above-mentioned three motion vectors of x direction of principal axis is p_x=max ([V _{N-3_x}, V _{N-2_x}, V _{N-1_x}]), the minimum value on the x direction of principal axis in above-mentioned three motion vectors is q_x=min ([V _{N-3_x}, V _{N-2_x}, V _{N-1_x}]); According to objective function S _Min=| pre_vx-V _{N-3_x}|+| pre_vx-V _{N-2_x}|+| pre_vx-V _{N-1_x}|, when said objective function is got minimum value, obtain the vector value pre_vx of target frame on the x axle, pre_vx ∈ [q_x, p_x];

(g), choose target frame three motion vector V before _N-3, V _N-2, V _N-1, the motion vector of target frame is pre_v; The maximal value of setting above-mentioned three motion vectors of y direction of principal axis is p_y=max ([V _{N-3_y}, V _{N-2_y}, V _{N-1_y}]), the minimum value on the y direction of principal axis in above-mentioned three motion vectors is q_y=min ([V _{N-3_y}, V _{N-2_y}, V _{N-1_y}]); According to objective function S _Min=| pre_vy-V _{N-3_y}|+| pre_vy-V _{N-2_y}|+| pre_vy-V _{N-1_y}|, when said objective function is got minimum value, obtain the vector value pre_vy of target frame on the y axle, pre_vy ∈ [q_y, p_y];

(h), obtain the motion vector pre_v of target frame, and, accomplish the prediction of target frame motion vector according to the motion vector modified value α that step e obtains according to step (f) and step (g).

2. photoconductive shipping dynamic vector Forecasting Methodology according to claim 1 is characterized in that: in said step (f) and the step (g), work as V _N-3=V _N-1The time, pre_v=V _N-1

3. photoconductive shipping dynamic vector Forecasting Methodology according to claim 1 is characterized in that: in said step (f) and the step (g), work as V _N-3≠ V _N-1, and work as V _N-3=V _N-2The time, pre_v=V _N-2

4. photoconductive shipping dynamic vector Forecasting Methodology according to claim 1 is characterized in that: in said step (f) and the step (g), work as V _N-3≠ V _N-1, and work as V _N-3≠ V _N-2The time, pre_v=V _N-2

Images