CN103411535B - A kind of Changeable weight picture point localization method for retro-reflective target - Google Patents

Abstract

The invention provides a kind of Changeable weight picture point localization method for retro-reflective target, described method comprises the steps: a) to arrange multiple object space monumented point; B) at different erect-position, described multiple object space monumented point is taken, obtain the gray-scale value at the multiple location point places on imaging facula; C) the second weight factor β relevant with ground unrest is obtained; D) the first weight factor α relevant to measurement field environmental variance is obtained; E) according to the first obtained weight factor α and the second weight factor β, the dot center coordinate (x of measured oval picture point is calculated ₀, y ₀).Center positioning method for retro-reflective target according to the present invention is more general compared with previous methods, goes for different site environments, eliminates the impact that site environment causes as much as possible, improve the positioning precision of retro-reflective target dot center.

Description

A kind of Changeable weight picture point localization method for retro-reflective target

Technical field

The present invention relates to photogrammetric technology field.Specifically, the present invention relates to a kind of Changeable weight picture point localization method for retro-reflective target (Retro-Reflective Targets is called for short RRT).

Background technology

The photogrammetric two dimensional image utilizing photography to obtain measured object, obtains the three-dimensional information of measured object.20th century the mid-80, close-range photogrammetry succeeds in commercial Application, it utilizes large format grid camera, many picture convergences calculate, digital picture scans, process, the technology such as to compare, and provides up to 1:500, the relative accuracy of 000, the precision of transit can be exceeded to the measurement of large sized object (e.g., diameter >10 rice).The mid-90, after combining digital sensor (CCD & COMS) technology, close-range photogrammetry is called as off-line Digital Photogrammetric System.By high resolving power slr camera, retro-reflective target point, and the technology of pixel location, it can provide very high precision level.Its typical measuring accuracy is 1:100,000 to 1:200,000 relative accuracy (1 times of standard deviation), the former corresponding 10 meters of target sizes are issued to the absolute precision of 0.1mm, and the absolute precision of general linear measure longimetry characterizes the relative accuracy of three-dimensional reconstruction, general lower than spot placement accuracy 2 to 3 times (such as, 0.05mm being about to the object of 2m).

Retro-reflective target is widely used a kind of artificial target in the high-precision industrial photogrammetry of recent domestic, and it adopts the special glass microballoon material structure with high index of refraction.Ad-hoc location use light source (such as flashlamp) irradiate time, RRT can by incident light according to original multipath tolerant to light source place, as shown in Figure 1.It has high reflectivity, is generally the hundred times even thousands of times of common sign back rate, to reach the accurate bianry image that impact point and background " are separated ", contributes to the accurate location of impact point and the extraction of unique point and identification.

After obtaining accurate bianry image, can position RRT.Usual RRT is a circle having a certain size, the coordinate position of its home position namely representated by this RRT.For extracting this position, there is diverse ways for different situations.Be the epoch of film at camera photo-sensitive cell, under normally film being placed in high magnification magnifier, obtain one amplify after RRT image, then found the center of RRT by method of geometry.

Along with the development of modern science and technology and manufacturing technology, especially be development and the develop rapidly of the solid state image sensor technology of important representative with CCD and COMS, adopt digital camera to obtain digitized video more and more universal, substantially instead of traditional film type camera.The unlike film camera of digital camera, sensing element can be taken out amplification and RRT is positioned, and due to the restriction of CCD and COMS sensor development level, its spatial resolution can not show a candle to the height of film.Circular retro-reflective target is generally ellipse after lens imaging, and the schematic diagram of imaging facula as shown in Figure 2.

By the restriction of optical system and imageing sensor manufacturing process, the resolution relative accuracy of image is no more than 1:10,000, for reaching higher relative accuracy, other is accurately located to need to carry out sub-pixel to light echo reflection spot, namely needs to position the center of imaging ellipse light spot.In order to reach the hi-Fix to elliptical center, usually there are two kinds of methods.One class methods ask for RRT edge, then simulates elliptic equation, thus try to achieve RRT elliptical point center; Another kind of method is the half-tone information based on RRT, tries to achieve RRT elliptical point center by centroid method.Wherein the accuracy requirement of extracting of edge fitting method edge point is very high, when monumented point size is less, and can because edge extracting error cause elliptical point center to be had a strong impact on.And the precision extracted based on the centroid method edge point of gray scale is insensitive, under the monumented point of different size, extraction accuracy change is little, and its shortcoming is easily subject to the uneven impact of intensity profile.

Conventional sub-pixel positioning algorithm has fitting process, digital correlation, centroid method.The prerequisite of fitting process is used to be that target property meets functional form that is known or supposition, so the sub-pixel precision that general more difficult acquisition is very high.Digital correlation has that principle is simple, strong adaptability and precision advantages of higher, but to rolling target or to rotate in abandoned target localization and seldom use relevant method, because this makes the selection of template be difficult to realize, and in close-range photogrammetry, RRT is always inevitably with rotation, and therefore correlation method can not be used for RRT dot center location.The grey scale centre of gravity method, square weighting method etc. of grey scale centre of gravity method, band threshold value are the sub-pixel recognition formed based on centroid method, the advantage of these algorithms is to take full advantage of the gray-scale value of every bit in RRT, usually can obtain the sub-pixel precision higher than centroid method, precision is at about 1/20px to 1/50 pixel.But the experimental result according to different document shows, the point center distill precision quality of these methods is not absolute, but under different occasions, obtains different point center distill precision.Because these methods are subject to the impact of Different field environment, thus show inconsistent performance level, therefore when environmental change, it is very inconvenient to use.

Therefore, need a kind of more general center positioning method for retro-reflective target, go for different site environments, eliminate the impact that site environment causes as much as possible, to improve the positioning precision of retro-reflective target dot center.

Summary of the invention

The object of the present invention is to provide a kind of variable weighting centroid method for retro-reflective target location, to improve the positioning precision of dot center.

According to an aspect of the present invention, provide a kind of Changeable weight picture point localization method for retro-reflective target, described method comprises the steps: a) to arrange multiple object space monumented point; B) at different erect-position, described multiple object space monumented point is taken, obtain the gray-scale value at the multiple location point places on imaging facula; C) the second weight factor β relevant with ground unrest is obtained; D) the first weight factor α relevant to measurement field environmental variance is obtained; E) according to the first obtained weight factor α and the second weight factor β, the dot center coordinate (x of measured oval picture point is calculated based on following formula ₀, y ₀),

$x_0=\frac{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{x}_i{(g(x_i,y_j)-\mathrm{\β})}^{\mathrm{\α}}}{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{(g(x_i,y_j)-\mathrm{\β})}^{\mathrm{\α}}}$

$y_0=\frac{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{y}_i{(g(x_i,y_j)-\mathrm{\β})}^{\mathrm{\α}}}{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{(g(x_i,y_j)-\mathrm{\β})}^{\mathrm{\α}}}$

Wherein x _iand y _jfor the coordinate (x of the specific location point on imaging facula _i, y _j), g (x _i, y _j) be (x _i, y _j) gray-scale value of place's imaging facula, [w ₁, w ₂] be the width range of imaging facula, [h ₁, h ₂] be the altitude range of imaging facula.

Preferably, described object space monumented point is the circular RRT mark utilizing retroreflecting material to make.

Preferably, the diameter range of described RRT mark is at 3-12mm.

Preferably, the diameter of described RRT mark is 6mm.

Preferably, described object space monumented point is evenly distributed in tested field.

Preferably, described second weight factor β is the average gray value of captured image.

Preferably, described second weight factor β is 0.

Preferably, [w ₁, w ₂] and [h ₁, h ₂] span is slightly larger than the width w of imaging facula ₀with height h ₀.

Preferably, in described step d), the computing method of the optimal value of the first weight factor α are as follows: d1) measure the coordinate obtaining each imaging point on captured picture, extract the two-dimensional points centre coordinate (x tried to achieve _i0, y _i0) as true value; D2) algorithm comprising described formula is used to obtain each point coordinate (x on picture _i, y _i) as measured value; D3) average of the residual error of described measured value and described true value is obtained; D4) getting α corresponding to minimum mean is optimal value.

Preferably, the value of described first weight factor α is integer.

Center positioning method for retro-reflective target according to the present invention is more general compared with previous methods, goes for different site environments, eliminates the impact that site environment causes as much as possible, improve the positioning precision of retro-reflective target dot center.

Should be appreciated that description and the follow-up detailed description of aforementioned cardinal principle are exemplary illustration and explanation, should not be used as the restriction to the claimed content of the present invention.

Accompanying drawing explanation

Describe above and other aspect of the present invention by reference to the accompanying drawings in detail, in accompanying drawing:

Fig. 1 shows the light path schematic diagram that RRT reflects the incident light when ad-hoc location uses light source irradiation;

Fig. 2 shows the schematic diagram of the imaging facula of circular retro-reflective target after lens imaging;

Fig. 3 shows the process flow diagram according to Changeable weight picture point localization method of the present invention;

Fig. 4 shows the experiment controlling filed schematic diagram of checking the inventive method;

Fig. 5 shows the picture point photo utilizing the experiment controlling filed of Fig. 4 to take;

Fig. 6 shows the relation curve of weight coefficient α and two-dimensional coordinate residual error in the plane of delineation.

Embodiment

The present invention proposes a kind of Changeable weight picture point localization method for retro-reflective target.By the relating environment values in conjunction with measurement field, by the correlative factor of measurement field, the such as distribution of angle, illumination and monumented point, and the ground unrest etc. of shooting is set as the weight factor that can change thus comprehensively considers, proposes a kind of localization method with the calculating retro-reflective target center of Changeable weight.Based on the corresponding weight of environment variable settings, can therefore be applicable to different measurement environment according to method of the present invention, thus obtain optimum weight coefficient, experiment proves to reach the positioning precision higher than prior art.

Fig. 3 shows the process flow diagram according to Changeable weight picture point localization method of the present invention.

In step 301, multiple object space monumented point is set.Object space monumented point is the circular RRT mark utilizing retroreflecting material to make, and the diameter range of RRT mark, at 3-12mm, is preferably about 6mm.Preferably, object space monumented point should be evenly distributed in tested field as far as possible.

In step 302, at different erect-position, the multiple object space monumented points set by be measured are taken, obtain the multiple location point (x on imaging facula _i, y _j) gray-scale value at place.Erect-position refers to device for shooting, such as residing during camera pictures taken position, and erect-position can by the coordinate (x of filming apparatus at global coordinate system, y, z) point to three corresponding rotation angle (θ, ω, κ) with the optical axis of filming apparatus to represent.The coordinate distribution of filming apparatus erect-position should be evenly distributed on to survey on the ball conical surface that object centers is the centre of sphere as far as possible, and cone angle, between 75 ° ~ 105 °, generally gets 90 °, and optical axis should point to the centre of sphere as far as possible.

In step 303, obtain the parameter of camera, namely obtain the second weight factor β relevant with ground unrest.Usually the average gray value of captured image can be used as the second weight factor β.In some occasion, such as when RRT size is very large and brightness uniformity when, also can ignore β.

In step 304, obtain the first weight factor α relevant to measurement field environmental variance, make (x ₀, y ₀) be dot center's coordinate of oval picture point measured by trying to achieve, (x ₀, y ₀) and relation derivation between the first weight factor α and the second weight factor β as follows:

First, make

$x_0=\frac{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{x}_{i}F(x_i,y_j,k_1,k_2,...k_n)}{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}F(x_i,y_j,k_1,k_2,...k_n)}$

$y_0=\frac{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{y}_{i}F(x_i,y_j,k_1,k_2,...k_n)}{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}F(x_i,y_j,k_1,k_2,...k_n)}---\left(1\right)$

Wherein, F (x _i, y _j, k ₁, k ₂... k _n) be and the specific location point (x on imaging facula _i, y _j) function that place's gray-scale value is relevant.K ₁, k ₂... k _nfor the coefficient of function, the scope asked for is x _i∈ [w ₁, w ₂], y _j∈ [h ₁, h ₂], wherein [w ₁, w ₂] be the width range of imaging facula, [h ₁, h ₂] be the altitude range of imaging facula.Preferably, [w ₁, w ₂] and [h ₁, h ₂] span is slightly larger than the width w of imaging facula ₀with height h ₀, generally get spot width absolute value, i.e. Shu w ₁-w ₂shu=w ₀+ 4, facular height absolute value, i.e. Shu h ₁-h ₂shu=h ₀the unit of+4(4 is pixel).

Make (x _i, y _j) place's gray-scale value is g (x _i, y _j), then can by the F (x in (1) formula _i, y _j, k ₁, k ₂... k _n) writing polynomial form, namely

$F(x_i,y_j,k_1,k_2,...k_n)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{k}_i\×g{(x_i,y_i)}^{n-i}---\left(2\right)$

When getting two rank after formula (2) right side is launched, have

F(x _i,y _j,k ₁,k ₂,…k _n)=k ₁·g(x _i,y _j) ²+k ₂·g(x _i,y _j) ¹+k ₃(3)

Be out of shape,

F(x _i,y _j,k ₁,k ₂,…k _n)=K ₁(g(x _i,y _j)+K ₂) ²+K ₃(4)

Wherein K ₂=k ₁, $K_2=\frac{k_2}{4k_1}$ $K_3=\frac{4k_1{k}_3-k_2^2}{4k_1}.$

From (1) formula, K ₁value does not affect result, therefore gets 1.K is obtained again from actual measurement experience ₂and K ₃substantially identical on the impact of dot center location, therefore for shortcut calculation can be merged, so have

F(x _i,y _j,k ₁,k ₂,…k _n)=(g(x _i,y _j)-β) ^α(5)

Wherein α is polynomial expression the highest item n-1, β=K ₂, α ∈ N ₊, β>=-g (x _i, y _j), namely

$x_0=\frac{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{x}_i{(g(x_i,y_j)-\mathrm{\β})}^{\mathrm{\α}}}{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{(g(x_i,y_j)-\mathrm{\β})}^{\mathrm{\α}}}$

$y_0=\frac{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{y}_i{(g(x_i,y_j)-\mathrm{\β})}^{\mathrm{\α}}}{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{(g(x_i,y_j)-\mathrm{\β})}^{\mathrm{\α}}}---\left(6\right)$

In step 305, according to the first obtained weight factor α and the second weight factor β, calculate the dot center coordinate (x of measured oval picture point based on formula (6) ₀, y ₀), namely complete the location of picture point.

First weight factor α according to the present invention reflects the various variablees relevant with measurement field environment, can select different α values for different environment fields, thus obtains the highest localization method of precision.The calculating of the optimal value of the first weight factor α can be in the following way.

The coordinate (such as utilizing VSTARS software) obtaining each imaging point on captured picture can be measured, extract the two-dimensional points centre coordinate (x tried to achieve _i0, y _i0) as true value; The algorithm routine that use comprises variable weight model (namely above-mentioned formula (6)) obtains each point coordinate (x on picture _i, y _i) as measured value; With the average of the residual error of measured value and true value as evaluation criterion, because α ∈ is Q, so residual error average along with α consecutive variations, and should must obtain minimum value at specific α place, this α value and optimal weights.

Not only can comprise the factor of environmental change according to formula of the present invention (6), namely reflect different measurement field situations by the value of α and β, can also compatible existing method for calculating and locating.Such as, when α=1, during β=0, formula (6) is traditional gray scale centroid method, that is,

$x_0=\frac{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{x}_{i}f(x_i,y_j)}{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}f(x_i,y_j)}$ $y_0=\frac{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{y}_{i}f(x_i,y_j)}{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}f(x_i,y_j)}---\left(7\right)$

Wherein f (x _i, y _j) be coordinate be (x _i, y _j) gray-scale value of place's point, the scope asked for is x _i∈ [w ₁, w ₂], y _j∈ [h ₁, h ₂], (x ₀, y ₀) be dot center's coordinate of trying to achieve.

When α=1, β=th(th represents threshold value) time, formula (6) is the gray scale centroid method of band threshold value, that is,

$x_0=\frac{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{x}_i[f(x_i,y_j)-\mathrm{th}]}{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}[f(x_i,y_j)-\mathrm{th}]}$ $y_0=\frac{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{y}_i[f(x_i,y_j)-\mathrm{th}]}{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}[f(x_i,y_j)-\mathrm{th}]}---\left(8\right)$

Wherein f (x _i, y _j) be coordinate be (x _i, y _j) gray-scale value of place's point, th is threshold constant.

When α=2, during β=0.Formula (6) is gray scale square weighting centroid method, namely

$x_0=\frac{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{x}_{i}f{(x_i,y_j)}^2}{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}f{(x_i,y_j)}^2}$ $y_0=\frac{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{y}_{i}f{(x_i,y_j)}^2}{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}f{(x_i,y_j)}^2}---\left(9\right)$

Wherein f (x _i, y _j) be coordinate be (x _i, y _j) gray-scale value of place's point, the scope asked for is x _i∈ [w ₁, w ₂], y _j∈ [h ₁, h ₂], square process is done to grey scale pixel value here.

As can be seen here, change environmentally can adapt to different situations according to localization method of the present invention, when RRT size is very large, when brightness uniformity, can β be ignored, make β=0, and get α=1; When the image that shooting obtains has certain background Gaussian noise, can set β is certain threshold value; And when RRT size is little, when the impact of pixel disc of confusion can not be ignored, the impact of pixel disc of confusion can be reflected by the size setting α, such as get α=2 or α=3 etc.

In above method, α value is integer, therefore corresponding different measurement environment.More preferably, under the measurement environment of reality more complicated and changeable, α value also can not be integer.α ∈ Q can be made.To different measurement environment, the optimal value of weight α is not fixed value, but determines along with the relating environment values of measurement field, such as angle, illumination, the distribution etc. of monumented point.Namely

$\left\{\begin{array}{cc}{x}_0=\frac{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{x}_i{(g(x_i,y_j)-\mathrm{\β})}^{\mathrm{\α}}}{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{(g(x_i,y_j)-\mathrm{\β})}^{\mathrm{\α}}}& y_0=\frac{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{y}_i{(g(x_i,y_i)-\mathrm{\β})}^{\mathrm{\α}}}{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{(g(x_i,y_j)-\mathrm{\β})}^{\mathrm{\α}}}\\ \mathrm{\α}=F(k_1,k_2\·\·\·,k_n)& \end{array}\right.---\left(10\right)$

Wherein, α ∈ Q, β>=-g (x _i, y _j), k ₁, k ₂, k _nfor the relating environment values of measurement field.Formula (10) namely reflects the variable picture point localization method of weight factor.

experimental result

1. data acquisition

Experimental verification is carried out to Changeable weight localization method of the present invention.As shown in Figure 4, its size is 4m*3m*1.5m to the schematic diagram of controlling filed.In the controlling filed shown in Fig. 4, arrange several RRT, be equally distributed as shown in Figure 4.Fig. 4 just schematically shows a part and layouts.The industrial digital up short coordinate measuring machine V-STARS E4X of GSI company of U.S. development is used to take.Filming apparatus adopts Nikon D2Xs, 1,200 ten thousand pixels, and resolution is 4288 × 2848, and pixel size is 6 μm, uses doughnut-shaped flash lamp to expose.Process software is V-STARS4.3, and its dot center's positioning precision can reach 1/50px.

With Nikon D2Xs camera at different erect-position shooting multiple pictures, amount to 67.The picture local that shooting obtains as shown in Figure 5.

2. data processing

Under the controlling filed environment shown in Fig. 4, owing to employing high-quality photographic goods, can control ground unrest is 0 substantially, then now β=0.(6) formula can abbreviation be like this

$x_0=\frac{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{x}_i{\left(g(x_i,y_j)\right)}^{\mathrm{\α}}}{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{\left(g(x_i,y_j)\right)}^{\mathrm{\α}}}$ $y_0=\frac{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{y}_i{\left(g(x_i,y_j)\right)}^{\mathrm{\α}}}{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{\left(g(x_i,y_j)\right)}^{\mathrm{\α}}}---\left(10\right)$

Such as, because optimal weights factor alpha is determined by some correlative factors of measurement field, the distribution of angle, illumination, monumented point, i.e. α=F (k ₁, k ₂..., k _n)

Funtcional relationship F is wherein also indefinite at present, therefore the optimal weights factor alpha under using the method compared to find present test field environment.When asking optimal weights, use following methods:

Use VSTARS software to obtain each point coordinate on picture, extract the two-dimensional points centre coordinate (x that VSTARS tries to achieve _i0, y _i0) as true value; Use the algorithm routine comprising variable weight model, obtain each point coordinate (x on picture _i, y _i) as measured value; With the average of the residual error of measured value and true value as evaluation criterion, because α ∈ is Q, so residual error average should along with α consecutive variations.And minimum value must be obtained at specific α place, this α value and optimal weights.Data processed result as shown in Figure 6.

As seen from Figure 6, residual error average with weight coefficient α change in [1.0,2.0] are interval, and obtain minimum value when α is about 1.6.Compare with the gray scale square weighting centroid method of α=2.0 with traditional centroid method of α=1.0, improve at least 25%.The optimal weights α tried to achieve by said method, when getting optimal weights, evaluation index is all better than classic method, illustrates that the precision using variable weight gray scale centroid method to extract dot center is better than classic method.

By above-mentioned experimental verification according to Changeable weight localization method of the present invention, its point center distill precision is better than traditional centroid method and gray scale square weighting centroid method, by calculating the standard deviation of residual error as the standard evaluating localization method, residual error wherein refers to that method according to the present invention tries to achieve dot center coordinate Coord (u, v) and coordinate true value Coord0 (u ₀, v ₀) between the mould of difference || Coord-Coord0||.Experimental result surface, the standard deviation of residual error at least can improve more than 25%.Therefore, use according to after variable weight model, the method that dot center asks for precision relatively traditional is improved.According to variable weight model, optimal weights α is not fixed value, is and changes with different measurement field.Concrete optimal weights factor alpha is determined by some correlative factors of measurement field, the such as distribution of angle, illumination, monumented point.

Should be understood that, under the prerequisite not deviating from the spirit of the present invention described in appended claim, the present invention can have multiple combination, correction, change and alternative.The foregoing is only the embodiment in the present invention; but protection scope of the present invention is not limited thereto; any people being familiar with this technology is in the technical scope disclosed by the present invention; the conversion or replacement expected can be understood; all should be encompassed in and of the present inventionly comprise within scope; therefore, protection scope of the present invention should be as the criterion with the protection domain of claims.In conjunction with the explanation of the present invention disclosed here and practice, other embodiments of the present invention are all apparent for those skilled in the art.Illustrate and embodiment be only considered to exemplary, true scope of the present invention and purport limited by claim.

Claims (7)

1., for a Changeable weight picture point localization method for retro-reflective target, described method comprises the steps:

A) multiple object space monumented point is set;

B) at different erect-position, described multiple object space monumented point is taken, obtain the gray-scale value at the multiple location point places on imaging facula;

C) obtain the second weight factor β relevant with ground unrest, wherein said second weight factor β is the average gray value of described imaging facula, or described second weight factor β is 0;

D) obtain the first weight factor α relevant to measurement field environmental variance, the computing method of the optimal value of described first weight factor α are as follows:

D1) measure the coordinate obtaining described multiple location point on imaging facula, extract the two-dimensional points centre coordinate (x tried to achieve _i0, y _i0) as true value;

D2) algorithm comprising following formula is used to obtain described multiple location point coordinate (x on imaging facula _i, y _j) as measured value, described formula is:

$x_{i0}=\frac{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{x}_i{(g(x_i,y_j)-\mathrm{\β})}^{\mathrm{\α}}}{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{(g(x_i,y_j)-\mathrm{\β})}^{\mathrm{\α}}}$

$y_{i0}=\frac{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{y}_i{(g(x_i,y_j)-\mathrm{\β})}^{\mathrm{\α}}}{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{(g(x_i,y_j)-\mathrm{\β})}^{\mathrm{\α}}};$

D3) average of the residual error of described measured value and described true value is obtained;

D4) getting α corresponding to minimum mean is optimal value;

E) according to the first obtained weight factor α and the second weight factor β, based on the dot center coordinate (x of oval picture point measured by following formulae discovery ₀, y ₀),

$x_0=\frac{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{x}_i{(g(x_i,y_j)-\mathrm{\β})}^{\mathrm{\α}}}{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{(g(x_i,y_j)-\mathrm{\β})}^{\mathrm{\α}}}$

$y_0=\frac{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{y}_i{(g(x_i,y_j)-\mathrm{\β})}^{\mathrm{\α}}}{\underset{i=h_1}{\overset{h_2}{\mathrm{\Σ}}}\underset{j=w_1}{\overset{w_2}{\mathrm{\Σ}}}{(g(x_i,y_j)-\mathrm{\β})}^{\mathrm{\α}}}$

Wherein x _iand y _jfor the coordinate (x of the location point on imaging facula _i, y _j), g (x _i, y _j) be (x _i, y _j) gray-scale value of place's imaging facula, [w ₁, w ₂] be the width range of imaging facula, [h ₁, h ₂] be the altitude range of imaging facula.

2. picture point localization method as claimed in claim 1, wherein said object space monumented point is the circular RRT mark utilizing retroreflecting material to make.

3. picture point localization method as claimed in claim 2, the diameter range of wherein said RRT mark is at 3-12mm.

4. picture point localization method as claimed in claim 3, the diameter of wherein said RRT mark is 6mm.

5. picture point localization method as claimed in claim 1, wherein said object space monumented point is evenly distributed in tested field.

6. picture point localization method, wherein [w as claimed in claim 1 ₁, w ₂] and [h ₁, h ₂] span is respectively slightly larger than the width w of imaging facula ₀with height h ₀.

7. picture point localization method as claimed in claim 1, the value of wherein said first weight factor α is integer.