CN105825470A - Fisheye image correction method base on point cloud image - Google Patents

Abstract

Provided in the invention is a fisheye image correction method base on a point cloud image. The method comprises: S1, cylindrical projection conversion is carried out on a fisheye image by using a projection center of a fisheye lens as a viewpoint, projection on a cylindrical surface is carried out, and then the projection image is unfolded to form a fisheye panorama image; S2, cylindrical projection conversion is carried out on point cloud data by using a projection center of a fisheye lens as a viewpoint, projection on a cylindrical surface is carried out, and then the projection image is unfolded to form a point cloud distance image; S3, edge detection and linear feature extraction are carried out on the fisheye panorama image; S4, edge detection and linear feature extraction are carried out on the point cloud distance image; and S5, the fisheye panorama image is corrected on the point cloud distance image based on affine transformation. According to the invention, with a point-cloud-image-based affine transformation model, secondary correction is carried out on the fisheye image after cylindrical projection conversion correction; and the corrected fisheye image can be used for texture mapping on a three-dimensional model. The calculation load is small and the feasibility of space measurement is high.

Description

Flake image rectification method based on a cloud image

Technical field

The present invention relates to image processing field, be specifically related to a kind of flake image rectification method based on a cloud image.

Background technology

Fish eye lens uses the hemisphere face camera lens of nonlinear organization, it it is a kind of extreme wide-angle lens, focal length is the most rounded at the image of about 8mm, visual angle is 180 °, minority even can reach 220 °, have only to two width fisheye photos in theory and just can complete the splicing of a width full-view image, thus greatly reduce shooting on the spot and the workload of follow-up photo process.Owing to fish eye lens uses the hemisphere face camera lens of nonlinear organization, therefore what flake image described is the scene information in half spherical space, and in the way of non-linear projection, record is in 2D plane, there is the most serious deformation, for beneficially eye-observation, it is necessary to it is corrected.

The bearing calibration of flake deformation of image mainly has two kinds:

1, the method for two dimensional surface conversion is directly used to be corrected, the method directly sets up respective coordinates transformation relation between flake image and correcting image, then by flake image projecting to correcting image, the most again to not having integer pixel values to carry out pixel interpolation processing.What this method was conventional is spherical coordinate orientation and multinomial coordinate transform etc..Spherical coordinate orientation method is the meridian analogizing on tellurion by flake image, the abscissa in the image after calibration of all pixels on each meridian that i.e. falls is the most identical, therefore can set up projection relation therebetween according to distortion ratio, flake image is realized correction.Multinomial coordinate transform is it is generally required to the multinomial of 5 even more high orders, and needs multiple control point, and therefore amount of calculation is huge, reasonably chooses control point simultaneously and also is difficult to hold.

2, the mode taking three dimensions flake image to convert is corrected, and conventional mainly have projection transform and fish eye lens are demarcated.Two dimensional surface imaging point each on flake image is mapped on spherical imaging face formation three-dimensional sphere imaging point by projection transform algorithm exactly, project to be constituted on image plane the two dimensional surface point of correcting image with the method for perspective projection the most again, it is the three-dimensional vector of the incident ray being gone out correspondence by imaging model from flake image pixel point inverse in fact, thus realizes correction；Fish eye lens calibration algorithm, initially sets up fish eye lens distorted pattern, then designs the distortion factor of panorama picture of fisheye lens, then is solved fish-eye inside and outside portion parameter by experiment, thus realizes the accurate correction of flake image.But, flake image is due to by the exterior orientation factor of fisheye camera and imaging rough error with affected, the mode taking three dimensions flake image to convert is corrected simply meeting the storage optimization demand of full-view image, cannot possess the feasibility of spatial measurement completely.

Summary of the invention

It is an object of the invention to, for problems of the prior art, it is provided that a kind of flake image rectification method based on a cloud image, its amount of calculation is less and possesses the feasibility of spatial measurement.

For achieving the above object, the present invention is by the following technical solutions:

A kind of flake image rectification method based on a cloud image, comprises the following steps:

S1, with fish-eye projection centre as viewpoint, flake image is carried out cylindrical surface projecting conversion, project on cylinder and launch formed flake full-view image；

S2, with fish-eye projection centre as viewpoint, cloud data is carried out cylindrical surface projecting conversion, project on cylinder and launch formed some a cloud range image；

S3, flake full-view image is carried out rim detection and extraction of straight line；

S4, a cloud range image is carried out rim detection and extraction of straight line；

S5, by affine transformation, flake full-view image is corrected on a cloud range image.

Further, in described S1, use following steps that flake image carries out cylindrical surface projecting conversion:

S101, with fish-eye projection centre as initial point, set up with initial point as the centre of sphere, with the imaging plane being parallel to flake image and the plane half spherical model as cross section, with fish-eye focal length as radius crossing initial point；

S102, axis direction along half spherical model, project to the imaging point on flake image on the hemisphere face of half spherical model；

S103, set up the vertical direction straight line cylinder model as axle center, with fish-eye focal length as radius to cross initial point；Subpoint on hemisphere face described in S102 is projected on the cylinder of described cylinder model；

S104, cylinder unwrapping becoming two dimensional surface, all subpoints on cylinder i.e. constitute flake full-view image.

Further, in described S103, by the subpoint on hemisphere face described in S102 by etc. project on the cylinder of described cylinder model in the way of arc length mapping.

Further, in described S103, by the subpoint on hemisphere face described in S102 along its line direction projection with initial point to the cylinder of described cylinder model on.

Further, in described S2, use following steps that cloud data carries out cylindrical surface projecting conversion:

S201, with fish-eye projection centre as initial point, set up centered by initial point, with the vertical direction straight line cylinder model as axle center crossing initial point；

S202, the Laser Measuring point that need to be converted into image for each carry out cylindrical surface projecting conversion, are projected on the cylinder of the cylinder model described in S1, form cylindrical surface projecting point；

S203, cylinder unwrapping is become two dimensional surface, the coordinate of cylinder subpoint is defined by two dimensional surface；Calculate the distance value between each laser measuring point and initial point, the coordinate of distance value with corresponding cylindrical surface projecting point is associated record；

S204, using the distance value between laser measuring point and initial point as the gray value of corresponding cylindrical surface projecting point, the cylindrical surface projecting point on two dimensional surface described in S3 is processed into Raster Images, is a cloud range image.

Further, described S202 comprises the following steps:

S2021, set up with initial point as the centre of sphere, the Sphere Measurement Model with the radius of the cylinder model described in S201 as radius, along the line direction of laser measuring point and initial point by Laser Measuring spot projection to Sphere Measurement Model, form spherical projection point；

S2022, by the spherical projection point described in S2021 by etc. arc length mapping in the way of project on the cylinder of the cylinder model described in S1, formed cylindrical surface projecting point.

Further, described S202 includes: along the line direction of laser measuring point and initial point by Laser Measuring spot projection to the cylinder of cylinder model, form cylindrical surface projecting point.

Further, described S3 includes:

S301, flake full-view image is carried out greyscale transformation and contrast stretching；

S302, gray scale flake full-view image is carried out the disposal of gentle filter；

S303, employing Canny operator carry out rim detection to the image after S302 process；

S304, employing Hough transform carry out conllinear judgement to the edge point set detected in S303, obtain conllinear point set, then conllinear point set are carried out extraction of straight line.

Further, described S4 includes:

S401, a cloud range image is carried out the disposal of gentle filter；

S402, employing Canny operator carry out rim detection to the image after S401 process；

S403, employing Hough transform carry out conllinear judgement to the edge point set detected in S402, obtain conllinear point set, then conllinear point set are carried out extraction of straight line.

Further, described S5 includes:

S501, to flake full-view image and some cloud range image carry out affine transformation, obtain registrate transfer function；

S502, using the characteristic curve that extracts in S3 and S4 as registration primitive, retrain according to the conllinear between characteristic curve and obtain flake full-view image and the similarity measure of some cloud range image, and obtain conllinear constraint equation；

S503, utilize the conllinear constraint equation obtained in S502, calculate the registration conversion parameter in registration transfer function, flake full-view image is corrected on a cloud range image.

A kind of based on a cloud image the flake image rectification method that the present invention provides, with fish-eye projection centre as viewpoint, respectively flake image and cloud data are carried out cylindrical surface projecting conversion, generate flake full-view image and some cloud range image, after extracting the linear feature of two kinds of images, by affine transformation, flake full-view image is corrected on a cloud range image.Owing to flake image by exterior orientation factor and the imaging rough error of fisheye camera and is affected, the single cylindrical surface projecting conversion correction in the past used simply meets the storage optimization demand of full-view image, cannot possess the feasibility of spatial measurement completely；And the present invention uses the flake image after correcting cylinder projective transformation based on an affine Transform Model of cloud image to carry out secondary correction, flake image after correction can carry out texture mapping directly to threedimensional model, and its amount of calculation is less and possesses the feasibility of spatial measurement.

Accompanying drawing explanation

Fig. 1 is the flow chart of a kind of based on a cloud image the flake image rectification method that the embodiment of the present invention one provides；

Fig. 2 is the projection process schematic diagram of fish eye lens macrura reevesii eye shadow picture upward in the embodiment of the present invention one；

Fig. 3 be in the embodiment of the present invention one fish eye lens towards the projection process schematic diagram of horizontal direction macrura reevesii eye shadow picture；

Fig. 4 is the projection process schematic diagram of the embodiment of the present invention one point cloud data；

Fig. 5 is flake full-view image and the affine transformation schematic diagram putting cloud range image in the embodiment of the present invention one；

Fig. 6 is the projection process schematic diagram of fish eye lens macrura reevesii eye shadow picture upward in the embodiment of the present invention two；

Fig. 7 be in the embodiment of the present invention two fish eye lens towards the projection process schematic diagram of horizontal direction macrura reevesii eye shadow picture；

Fig. 8 is the projection process schematic diagram of the embodiment of the present invention two point cloud data.

Detailed description of the invention

Below in conjunction with accompanying drawing and specific embodiment, technical scheme is described in detail.

Embodiment one

Embodiments provide a kind of flake image rectification method based on a cloud image, as it is shown in figure 1, it comprises the following steps:

S1, with fish-eye projection centre as viewpoint, flake image is carried out cylindrical surface projecting conversion, project on cylinder and launch formed flake full-view image；

S2, with fish-eye projection centre as viewpoint, cloud data is carried out cylindrical surface projecting conversion, project on cylinder and launch formed some a cloud range image；

S3, flake full-view image is carried out rim detection and extraction of straight line；

S4, a cloud range image is carried out rim detection and extraction of straight line；

S5, by affine transformation, flake full-view image is corrected on a cloud range image.

In the present embodiment, use the mode of cylinder equidistant projection conversion that flake image and cloud data are carried out projection process.

Specifically, described S1 includes:

S101, with fish-eye projection centre as initial point, set up with initial point as the centre of sphere, with the imaging plane being parallel to flake image and the plane half spherical model as cross section, with fish-eye focal length as radius crossing initial point；

S102, axis direction along half spherical model, project to the imaging point on flake image on the hemisphere face of half spherical model；

S103, set up the vertical direction straight line cylinder model as axle center, with fish-eye focal length as radius to cross initial point；By the subpoint on hemisphere face described in S102 by etc. project on the cylinder of described cylinder model in the way of arc length mapping；

S104, cylinder unwrapping becoming two dimensional surface, all subpoints on cylinder i.e. constitute flake full-view image.

The fish-eye imaging model used in the present invention is spherical perspective model, flake image can be regarded as being imaged on a centre of sphere be O, radius be on the hemisphere face of R.The 3-dimensional image point that each two-dimensional image planar point on flake image is mapped on hemisphere face by cylindrical surface projecting conversion exactly, i.e. for each pixel on flake image, a three-dimensional vector light of projection centre from photographing unit can be reverted to, namely incident ray, thus complete from image plane to hemispherical coordinate transform；Then according to hemisphere face radius, set up a circumscribed cylinder, just will can complete projective transformation in 3D vector light backprojection to cylinder.

Specifically, fish eye lens generate flake image time typically have two kinds towards, a kind of be camera lens upward, a kind of is camera lens level, and the difference of lens direction can cause the difference of projective transformation formula.

When fish eye lens upward time, as in figure 2 it is shown, with O as initial point, with the axle center of cylinder model as z-axis, with the cross section of half spherical model for xOy plane, set up three-dimensional cartesian coordinate system O-xyz.Equidistant imaging model according to flake calculates any point image coordinate A (x, y) the projection coordinate A'(θ in unit sphere on flake image₁,ξ₁), i.e. calculating incident ray is relative to the azimuth of camera imaging coordinate system, if unit sphere radius is focal distance f, their relation such as following formula:

$\left\{\begin{array}{c}x=f\×{\mathrm{sin\θ}}_1\×{\mathrm{sin\ξ}}_1\\ y=f\×{\mathrm{sin\θ}}_1\×{\mathrm{cos\ξ}}_1\end{array}\right.$

Wherein, θ₁For the angle of OA ' Yu z-axis, ξ₁Angle for OA Yu y-axis.

With the intersection points B of y-axis positive direction and the cylinder of cylinder model as initial point, on cylinder, y-axis positive direction is i direction of principal axis to the circumferencial direction of x-axis positive direction, axial for j direction of principal axis with cylinder, sets up two-dimensional direct angle coordinate system B-ij on cylinder；In flake image, picture point A is through sphere corresponding point A'(θ₁,ξ₁) reprojection is launched to cylinder tailing edge bus BD, is A in image space iBj corresponding point " (i, j), taking focal distance f is cylindrical radius.Fish eye lens upward time, camera lens primary optical axis is perpendicular to ground, and i, j are represented by formula:

$\left\{\begin{array}{c}i=f\×{\mathrm{\ξ}}_1\\ j=f\×(\frac{\mathrm{\π}}{2}-{\mathrm{\θ}}_1)\end{array}\right.$

Thus directly by spherical coordinate (θ₁,ξ₁) A on cylinder can be set up " (i, j) with flake image coordinate A (x, y) between corresponding relation.

When fish eye lens is towards horizontal direction, as it is shown on figure 3, with O as initial point, with the axle center of cylinder model as z-axis, with the axle center of half spherical model as x-axis, set up three-dimensional cartesian coordinate system O-xyz.The flake image conversion of camera lens level and camera lens upward different, incident ray is relative to the azimuth (θ of camera imaging coordinate system₂,ξ₂) it is no longer the azimuth (α, β) directly as three-dimensional cartesian coordinate system O-xyz.However, it is possible to set up cylindrical surface projecting coordinate (α, β) and incident ray spherical coordinate (θ₂,ξ₂) relation such as following formula:

$\left\{\begin{array}{c}{\mathrm{sin\θ}}_2=sin\mathrm{\α}\×sin\mathrm{\β}\\ {\mathrm{sin\ξ}}_2=\frac{sin\mathrm{\α}\×sin\mathrm{\β}}{\sqrt{{\sin }^2\mathrm{\α}\×{\sin }^2\mathrm{\β}+{\cos }^2\mathrm{\α}}}\\ {\mathrm{cos\ξ}}_2=\frac{cos\mathrm{\α}}{\sqrt{{\sin }^2\mathrm{\α}\×{\sin }^2\mathrm{\β}+{\cos }^2\mathrm{\α}}}\end{array}\right.$

Wherein, θ₂For the angle of OA ' Yu x-axis, ξ₂For the angle of OA Yu z-axis, α is the angle of OA ' and xOy plane, and β is the OA ' projection in xOy plane and the angle between y-axis.

Further, with the intersection points B of y-axis positive direction and the cylinder of cylinder model as initial point, on cylinder, y-axis positive direction is i direction of principal axis to the circumferencial direction of x-axis positive direction, axial for j direction of principal axis with cylinder, sets up two-dimensional direct angle coordinate system B-ij on cylinder；A is obtained according to the arc length such as inclined angle alpha, level orientation angle beta are carried out on cylinder model projection " (i, j):

$\left\{\begin{array}{c}i=f\×\mathrm{\β}\\ j=f\×\mathrm{\α}\end{array}\right.$

Thus by inclined angle alpha, level orientation angle beta and spherical coordinate A'(θ_2,ξ₂) A on cylinder can be set up " (i, corresponding relation j) and between flake image coordinate A.

Further, described S2 includes:

S201, with fish-eye projection centre as initial point, set up centered by initial point, with the vertical direction straight line cylinder model as axle center crossing initial point；

S202, the Laser Measuring point that need to be converted into image for each carry out cylindrical surface projecting conversion, are projected on the cylinder of the cylinder model described in S1, form cylindrical surface projecting point；

S203, cylinder unwrapping is become two dimensional surface, the coordinate of cylinder subpoint is defined by two dimensional surface；Calculate the distance value between each laser measuring point and initial point, the coordinate of distance value with corresponding cylindrical surface projecting point is associated record；

S204, using the distance value between laser measuring point and initial point as the gray value of corresponding cylindrical surface projecting point, the cylindrical surface projecting point on two dimensional surface described in S3 is processed into Raster Images, is a cloud range image.

Wherein, described S202 comprises the following steps:

S2021, set up with initial point as the centre of sphere, the Sphere Measurement Model with the radius of the cylinder model described in S201 as radius, along the line direction of laser measuring point and initial point by Laser Measuring spot projection to Sphere Measurement Model, form spherical projection point；

S2022, by the spherical projection point described in S2021 by etc. arc length mapping in the way of project on the cylinder of the cylinder model described in S1, formed cylindrical surface projecting point.

See Fig. 4, with fish-eye projection centre O (x₀, y₀, z₀) it is initial point, set up centered by initial point O, with R as radius, with the vertical direction straight line cylinder model as axle center crossing initial point O；And with projection centre O (x₀, y₀, z₀) it is initial point, with the axle center of cylinder model as z-axis, set up three-dimensional cartesian coordinate system O-xyz；For laser measuring point A (x₁, y₁, z₁), set up with initial point O as the centre of sphere, the Sphere Measurement Model with R as radius, along AO direction, laser measuring point A is projected on Sphere Measurement Model, form spherical projection point Wherein, θ₃For the angle of the projection in xOy plane of the line of laser measuring point A and initial point O with y-axis positive direction,Line and the angle of xOy plane for laser measuring point A and initial point O.With Sphere Measurement Model as reference, θ₃It also is understood as the spherical projection point A ' longitude on Sphere Measurement Model,It also is understood as the spherical projection point A ' latitude on Sphere Measurement Model.

Specifically, θ₃Computational methods as follows:

First, it is determined that laser measuring point A subpoint in xOy plane is relative to the quadrant residing for O-xy coordinate system；

Then according to equation below calculates:

$\left\{\begin{array}{cc}{\mathrm{\θ}}_3=arctan\left|\frac{x_1-x_0}{y_1-y_0}\right|+\frac{\mathrm{\π}k}{2},& k=(0,2)\\ {\mathrm{\θ}}_3=arctan\left|\frac{y_1-y_0}{x_1-x_0}\right|+\frac{\mathrm{\π}k}{2},& k=(1,3)\end{array}\right.$

Wherein, when laser measuring point A subpoint in xOy plane respectively falls in first, second, third and fourth quadrant of O-xy coordinate system, k value takes k=0 respectively, and 1,2,3.

Computing formula the most as follows:

Laser measuring point A projection on Sphere Measurement Model is obtained according to above-mentioned methodAfter, by spherical projection pointBy etc. project on the cylinder of cylinder model in the way of arc length mapping, form cylindrical surface projecting point A ".This process is i.e. one and is mapped according to the mode of arc length equably by equator to polar region, the most namely the ken in 180 ° of sphere latitude direction is carried out decile uniform projection again to cylinder according to latitude, and longitude aspect is still identical with cylinder equiangular transformation, as shown in Figure 4.

Finally, with the intersection points B of y-axis positive direction and the cylinder of cylinder model as initial point, on cylinder, y-axis positive direction is i direction of principal axis to the circumferencial direction of x-axis positive direction, axial for j direction of principal axis with cylinder, sets up two-dimensional direct angle coordinate system B-ij on cylinder；After then cylinder being launched into two dimensional surface along j axle (i.e. BD direction shown in Fig. 4), cylindrical surface projecting point A " coordinate on two dimensional surface (i, the value of distance value d j) and between laser measuring point A and initial point O is as follows:

Cloud data projects to after two dimensional surface with Raster Images record, and in Raster Images, the gray scale of each picture dot is determined by the distance value d of laser measuring point A to initial point O.Depending on the resolution of image is according to cloud data density, owing to the density of cloud data is the biggest, it is believed that there is not the probability in cavity.But there is the situation of the projection of multiple laser measuring point in certainly there is a picture dot, then the gray scale of this picture dot is just as the criterion with the laser measuring point closest to initial point, i.e. take the minima of distance value d in picture dot.

After cylindrical surface projecting conversion, there is bigger difference in flake full-view image and some cloud range image on vertical-horizontal proportion, but ensure that the invariance of vertical bar in real world the most well.Therefore, flake full-view image and the linear feature of some cloud range image are extracted the most respectively.

Further, described S3 includes:

S301, flake full-view image is carried out greyscale transformation and contrast stretching；

S302, gray scale flake full-view image is carried out the disposal of gentle filter；

S303, employing Canny operator carry out rim detection to the image after S302 process；

S304, employing Hough transform carry out conllinear judgement to the edge point set detected in S303, obtain conllinear point set, then conllinear point set are carried out extraction of straight line.

Described S4 includes:

S401, a cloud range image is carried out the disposal of gentle filter；

S402, employing Canny operator carry out rim detection to the image after S401 process；

S403, employing Hough transform carry out conllinear judgement to the edge point set detected in S402, obtain conllinear point set, then conllinear point set are carried out extraction of straight line.

Further, described S5 includes:

S501, to flake full-view image and some cloud range image carry out affine transformation, obtain registrate transfer function；

S502, using the characteristic curve that extracts in S3 and S4 as registration primitive, retrain according to the conllinear between characteristic curve and obtain flake full-view image and the similarity measure of some cloud range image, and obtain conllinear constraint equation；

S503, utilize the conllinear constraint equation obtained in S502, calculate the registration conversion parameter in registration transfer function, flake full-view image is corrected on a cloud range image.

Owing to the flake full-view image after projective transformation is satisfied by perspective imaging condition with some cloud range image, and there is identical viewpoint, simultaneously for meeting MIHT (ModifiedIteratedHoughTransform, the iteration Hough transform improved) algorithm too much characteristic unsuitable for parameter to be resolved, affine Transform Model can be used as registration transformation model.Affine transformation has good mathematical characteristic, and after affine transformation, straight line is still straight line, and have parallel relation also can keep constant.

If the coordinate of the point on flake full-view image is that (x, y), the coordinate of the point on some cloud range image is (x', y').Then according to affine Transform Model, the point coordinates on flake full-view image after projective transformation and the registration transfer function between the point coordinates on laser scanning point cloud range image be:

$\left[\begin{array}{c}{x}^{\′}\\ y^{\′}\end{array}\right]=\left[\begin{array}{c}{a}_0\\ b_0\end{array}\right]+\left[\begin{array}{cc}{a}_1& a_2\\ b_1& b_2\end{array}\right]\left[\begin{array}{c}x\\ y\end{array}\right]$

Formula has a₀、a₁、a₂、b₀、b₁、b₂Totally 6 parameters, a₀、b₀It is translation (Translation) parameter, a₁、a₂、b₁、b₂It is to rotate (Rotation), scaling (Scale), upset (Flip) and mistake to cut (Shear) parameter respectively, solves these 6 parameters and just can complete the correction of flake full-view image.

The characteristic curve of extraction of straight line can be obtained similarity measure as registration primitive, the conllinear constraint between characteristic curve.

As it is shown in figure 5, the Eigenvector L of flake full-view image₁After transforming to image coordinate XOY of cloud range image, end points P₁(x₁, y₁)、P₂(x₂, y₂) it is transformed to P₁’(x₁', y₁’)、P₂’(x₂', y₂’).Judge straightway L₁With straightway L₂The condition of conllinear is P₁’、P₂' distance feature line L₂Distance be all 0, i.e. n₁=n₂=0.If ρ be under image coordinate XOY initial point to straight line L₂Distance, θ is L₂Normal and the angle of X-axis, the most now the form of conllinear constraints dotted line antithesis is written as:

$\left\{\begin{array}{c}{n}_1={\mathrm{Px}}_1^{\′}cos\mathrm{\θ}+{\mathrm{Py}}_1^{\′}sin\mathrm{\θ}-\mathrm{\ρ}=0\\ n_2={\mathrm{Px}}_2^{\′}cos\mathrm{\θ}+{\mathrm{Py}}_2^{\′}sin\mathrm{\θ}-\mathrm{\ρ}=0\end{array}\right.$

From the foregoing, it will be observed that every a pair feature of the same name at least can obtain two conllinear constraint equations (can also be to carry out structure constraint equation according to any two points on characteristic curve) on two end points.Utilize n feature collinear relationship, 2n condition can be set up.Utilizing the conllinear constraint equation that these constraintss obtain, can calculate 6 registration conversion parameters in above-mentioned registration transfer function, formula is as follows:

$\left\{\begin{array}{c}{a}_0=\mathrm{\ρ}-\frac{y^{\′}\·sin\mathrm{\θ}}{\cos \mathrm{\θ}}-(a_1\·x+a_2\·y)\\ b_0=\mathrm{\ρ}-\frac{x^{\′}\·\cos \mathrm{\θ}}{\sin \mathrm{\θ}}-(b_1\·x+b_2\·y)\\ a_1=\[\mathrm{\ρ}-\frac{y^{\′}\·sin\mathrm{\θ}}{\cos \mathrm{\θ}}-(a_0+a_2\·y)\]/x\\ b_1=\[\mathrm{\ρ}-\frac{x^{\′}\·\cos \mathrm{\θ}}{\sin \mathrm{\θ}}-(b_0+b_2\·y)\]/x\\ a_2=\[\mathrm{\ρ}-\frac{y^{\′}\·\sin \mathrm{\θ}}{\cos \mathrm{\θ}}-(a_0+a_1\·x)\]/y\\ b_2=\[\mathrm{\ρ}-\frac{x^{\′}\·\cos \mathrm{\θ}}{\sin \mathrm{\θ}}-(b_0+b_1\·x)\]/y\end{array}\right.$

Embodiment two

A kind of based on a cloud image the flake image rectification method that the present embodiment provides is substantially the same with embodiment one, simply have employed the another kind of cylindrical surface projecting mapping mode different from embodiment one in S1 and S2 and flake image and cloud data are carried out projection process.Will be around S1 and S2 emphasis below and illustrate the difference of the present embodiment, S3 to S5 is identical with embodiment one, does not repeats them here.

Specifically, the present embodiment uses the mode of cylinder Conformal Projection Transformation that flake image and cloud data are carried out projection process in S1 and S2.

Described S1 includes:

S101, with fish-eye projection centre as initial point, set up with initial point as the centre of sphere, with the imaging plane being parallel to flake image and the plane half spherical model as cross section, with fish-eye focal length as radius crossing initial point；

S102, axis direction along half spherical model, project to the imaging point on flake image on the hemisphere face of half spherical model；

S103, set up the vertical direction straight line cylinder model as axle center, with fish-eye focal length as radius to cross initial point；By the subpoint on hemisphere face described in S102 along its line direction projection with initial point to the cylinder of described cylinder model on；

S104, cylinder unwrapping becoming two dimensional surface, all subpoints on cylinder i.e. constitute flake full-view image.

When fish eye lens upward time, as shown in Figure 6, with O as initial point, with the axle center of cylinder model as z-axis, with the cross section of half spherical model for xOy plane, set up three-dimensional cartesian coordinate system O-xyz.Equidistant imaging model according to flake calculates any point image coordinate A (x, y) the projection coordinate A'(θ in unit sphere on flake image₄,ξ₄), i.e. calculating incident ray is relative to the azimuth of camera imaging coordinate system, if unit sphere radius is focal distance f, their relation such as following formula:

$\left\{\begin{array}{c}x=f\×{\mathrm{sin\θ}}_4\×{\mathrm{sin\ξ}}_4\\ y=f\×{\mathrm{sin\θ}}_4\×{\mathrm{cos\ξ}}_4\end{array}\right.$

Wherein, θ₄For the angle of OA ' Yu z-axis, ξ₄Angle for OA Yu y-axis.

With the intersection points B of y-axis positive direction and the cylinder of cylinder model as initial point, on cylinder, y-axis positive direction is i direction of principal axis to the circumferencial direction of x-axis positive direction, axial for j direction of principal axis with cylinder, sets up two-dimensional direct angle coordinate system B-ij on cylinder；In flake image, picture point A is through sphere corresponding point A'(θ₄,ξ₄) reprojection is launched to cylinder tailing edge bus BD, is A in image space iBj corresponding point " (i, j), taking focal distance f is cylindrical radius.Fish eye lens upward time, when camera lens primary optical axis is perpendicular to ground, i, j are represented by formula:

$\left\{\begin{array}{c}i=f\×{\mathrm{\ξ}}_4\\ j=f/{\mathrm{tan\θ}}_4\end{array}\right.$

Thus directly by spherical coordinate (θ₄,ξ₄) A on cylinder can be set up " (i, j) with flake image coordinate A (x, y) between corresponding relation.

When fish eye lens is towards horizontal direction, as it is shown in fig. 7, with O as initial point, with the axle center of cylinder model as z-axis, with the axle center of half spherical model as x-axis, set up three-dimensional cartesian coordinate system O-xyz.The flake image conversion of camera lens level and camera lens upward different, incident ray is relative to the azimuth (θ of camera imaging coordinate system₅,ξ₅) it is no longer directly as inclination angle and horizontal azimuth (α, β).But also according to the principle of perspective, the subpoint A on cylinder " and sphere on subpoint A'(θ₄,ξ₄) inclined angle alpha be equal with level orientation angle beta, such that it is able to set up cylindrical surface projecting coordinate (α, β) and incident ray spherical coordinate (θ₅,ξ₅) relation such as following formula:

$\left\{\begin{array}{c}{\mathrm{sin\θ}}_5=sin\mathrm{\α}\×sin\mathrm{\β}\\ {\mathrm{sin\ξ}}_5=\frac{sin\mathrm{\α}\×sin\mathrm{\β}}{\sqrt{{\sin }^2\mathrm{\α}\×{\sin }^2\mathrm{\β}+{\cos }^2\mathrm{\α}}}\\ {\mathrm{cos\ξ}}_5=\frac{cos\mathrm{\α}}{\sqrt{{\sin }^2\mathrm{\α}\×{\sin }^2\mathrm{\β}+{\cos }^2\mathrm{\α}}}\end{array}\right.$

Wherein, θ₅For the angle of OA ' Yu x-axis, ξ₅For the angle of OA Yu z-axis, α is the angle of OA ' and xOy plane, and β is the OA ' projection in xOy plane and the angle between y-axis.

Further, with the intersection points B of y-axis positive direction and the cylinder of cylinder model as initial point, on cylinder, y-axis positive direction is i direction of principal axis to the circumferencial direction of x-axis positive direction, axial for j direction of principal axis with cylinder, sets up two-dimensional direct angle coordinate system B-ij on cylinder；On cylinder, carry out isometric projection according to inclined angle alpha, level orientation angle beta and obtain A " (i, j):

$\left\{\begin{array}{c}i=f\×\mathrm{\β}\\ j=f\×tan\mathrm{\α}\end{array}\right.$

Thus by inclined angle alpha, level orientation angle beta and spherical coordinate (θ₅,ξ₅) A on cylinder can be set up " (i, j) with flake image coordinate A (r, ξ₅Corresponding relation between).

Further, described S2 includes:

S201, with fish-eye projection centre as initial point, set up centered by initial point, with the vertical direction straight line cylinder model as axle center crossing initial point；

S202, the Laser Measuring point that need to be converted into image for each carry out cylindrical surface projecting conversion, by its along the line direction of laser measuring point and initial point by Laser Measuring spot projection to the cylinder of cylinder model, form cylindrical surface projecting point；

S203, cylinder unwrapping is become two dimensional surface, the coordinate of cylinder subpoint is defined by two dimensional surface；Calculate the distance value between each laser measuring point and initial point, the coordinate of distance value with corresponding cylindrical surface projecting point is associated record；

S204, using the distance value between laser measuring point and initial point as the gray value of corresponding cylindrical surface projecting point, the cylindrical surface projecting point on two dimensional surface described in S3 is processed into Raster Images, is a cloud range image.

See Fig. 8, with fish-eye projection centre O (x₀, y₀, z₀) it is initial point, set up centered by initial point O, with R as radius, with the vertical direction straight line cylinder model as axle center crossing initial point O；And with projection centre O (x₀, y₀, z₀) it is initial point, with the axle center of cylinder model as z-axis, set up three-dimensional cartesian coordinate system O-xyz；Along AO direction, laser measuring point A is projected on the cylinder of cylinder model, directly obtain cylindrical surface projecting point A ".

Then, with the intersection points B of y-axis positive direction and the cylinder of cylinder model as initial point, on cylinder, y-axis positive direction is i direction of principal axis to the circumferencial direction of x-axis positive direction, axial for j direction of principal axis with cylinder, sets up two-dimensional direct angle coordinate system B-ij on cylinder；After then cylinder being launched into two dimensional surface along j axle (i.e. BD direction shown in Fig. 8), cylindrical surface projecting point A " coordinate on two dimensional surface (i, the value of distance value d j) and between corresponding laser measuring point A and initial point O is as follows:

Wherein, R is the radius of cylinder model, θ₆For the angle of the projection in xOy plane of the line of laser measuring point A and initial point O with y-axis positive direction,Line and the angle of xOy plane for laser measuring point A and initial point O.

Further, θ₆Computational methods identical with embodiment one,Value be:

A kind of based on a cloud image the flake image rectification method that the present invention provides, with fish-eye projection centre as viewpoint, respectively flake image and cloud data are carried out cylindrical surface projecting conversion, generate flake full-view image and some cloud range image, after extracting the linear feature of two kinds of images, by affine transformation, flake full-view image is corrected on a cloud range image.Owing to flake image by exterior orientation factor and the imaging rough error of fisheye camera and is affected, the single cylindrical surface projecting conversion correction in the past used simply meets the storage optimization demand of full-view image, cannot possess the feasibility of spatial measurement completely；And the present invention uses the flake image after correcting cylinder projective transformation based on an affine Transform Model of cloud image to carry out secondary correction, flake image after correction can carry out texture mapping directly to threedimensional model, and its amount of calculation is less and possesses the feasibility of spatial measurement.

Embodiment described above only have expressed the several embodiments of the present invention, and it describes more concrete and detailed, but therefore can not be interpreted as the restriction to the scope of the claims of the present invention.It should be pointed out that, for the person of ordinary skill of the art, without departing from the inventive concept of the premise, it is also possible to make some deformation and improvement, these broadly fall into protection scope of the present invention.Therefore, the protection domain of patent of the present invention should be as the criterion with claims.

Claims (10)

1. a flake image rectification method based on a cloud image, it is characterised in that comprise the following steps:

S1, with fish-eye projection centre as viewpoint, flake image is carried out cylindrical surface projecting conversion, project on cylinder and launch formed flake full-view image；

S2, with fish-eye projection centre as viewpoint, cloud data is carried out cylindrical surface projecting conversion, project on cylinder and launch formed some a cloud range image；

S3, flake full-view image is carried out rim detection and extraction of straight line；

S4, a cloud range image is carried out rim detection and extraction of straight line；

S5, by affine transformation, flake full-view image is corrected on a cloud range image.

Flake image rectification method based on a cloud image the most according to claim 1, it is characterised in that in described S1, uses following steps that flake image carries out cylindrical surface projecting conversion:

S101, with fish-eye projection centre as initial point, set up with initial point as the centre of sphere, with the imaging plane being parallel to flake image and the plane half spherical model as cross section, with fish-eye focal length as radius crossing initial point；

S102, axis direction along half spherical model, project to the imaging point on flake image on the hemisphere face of half spherical model；

S103, set up the vertical direction straight line cylinder model as axle center, with fish-eye focal length as radius to cross initial point；Subpoint on hemisphere face described in S102 is projected on the cylinder of described cylinder model；

S104, cylinder unwrapping becoming two dimensional surface, all subpoints on cylinder i.e. constitute flake full-view image.

Flake image rectification method based on a cloud image the most according to claim 2, it is characterised in that in described S103, by the subpoint on hemisphere face described in S102 by etc. project on the cylinder of described cylinder model in the way of arc length mapping.

Flake image rectification method based on a cloud image the most according to claim 2, it is characterised in that in described S103, by the subpoint on hemisphere face described in S102 along its line direction projection with initial point to the cylinder of described cylinder model on.

Flake image rectification method based on a cloud image the most according to claim 1, it is characterised in that in described S2, uses following steps that cloud data carries out cylindrical surface projecting conversion:

S201, with fish-eye projection centre as initial point, set up centered by initial point, with the vertical direction straight line cylinder model as axle center crossing initial point；

S202, the Laser Measuring point that need to be converted into image for each carry out cylindrical surface projecting conversion, are projected on the cylinder of the cylinder model described in S1, form cylindrical surface projecting point；

S203, cylinder unwrapping is become two dimensional surface, the coordinate of cylinder subpoint is defined by two dimensional surface；Calculate the distance value between each laser measuring point and initial point, the coordinate of distance value with corresponding cylindrical surface projecting point is associated record；

S204, using the distance value between laser measuring point and initial point as the gray value of corresponding cylindrical surface projecting point, the cylindrical surface projecting point on two dimensional surface described in S3 is processed into Raster Images, is a cloud range image.

Flake image rectification method based on a cloud image the most according to claim 5, it is characterised in that described S202 comprises the following steps:

S2021, set up with initial point as the centre of sphere, the Sphere Measurement Model with the radius of the cylinder model described in S201 as radius, along the line direction of laser measuring point and initial point by Laser Measuring spot projection to Sphere Measurement Model, form spherical projection point；

S2022, by the spherical projection point described in S2021 by etc. arc length mapping in the way of project on the cylinder of the cylinder model described in S1, formed cylindrical surface projecting point.

Flake image rectification method based on a cloud image the most according to claim 5, it is characterised in that described S202 includes: along the line direction of laser measuring point and initial point by Laser Measuring spot projection to the cylinder of cylinder model, form cylindrical surface projecting point.

Flake image rectification method based on a cloud image the most according to claim 1, it is characterised in that described S3 includes:

S301, flake full-view image is carried out greyscale transformation and contrast stretching；

S302, gray scale flake full-view image is carried out the disposal of gentle filter；

S303, employing Canny operator carry out rim detection to the image after S302 process；

S304, employing Hough transform carry out conllinear judgement to the edge point set detected in S303, obtain conllinear point set, then conllinear point set are carried out extraction of straight line.

Flake image rectification method based on a cloud image the most according to claim 1, it is characterised in that described S4 includes:

S401, a cloud range image is carried out the disposal of gentle filter；

S402, employing Canny operator carry out rim detection to the image after S401 process；

S403, employing Hough transform carry out conllinear judgement to the edge point set detected in S402, obtain conllinear point set, then conllinear point set are carried out extraction of straight line.

Flake image rectification method based on a cloud image the most according to claim 1, it is characterised in that described S5 includes:

S501, to flake full-view image and some cloud range image carry out affine transformation, obtain registrate transfer function；

S502, using the characteristic curve that extracts in S3 and S4 as registration primitive, retrain according to the conllinear between characteristic curve and obtain flake full-view image and the similarity measure of some cloud range image, and obtain conllinear constraint equation；

S503, utilize the conllinear constraint equation obtained in S502, calculate the registration conversion parameter in registration transfer function, flake full-view image is corrected on a cloud range image.