CN103198481B - A kind of camera marking method - Google Patents

Abstract

The invention discloses a kind of camera marking method and realize system, belong to image procossing and computer vision field, the method comprises the following steps: in target plane sets two unique points; Obtain the image coordinate of the unique point on target, set up unique point three dimensional space coordinate (X _w, Y _w, Z _w) and its picture point coordinate (X _f, Y _f) between corresponding relation; This corresponding relation is updated in spatial mappings model; Use optimization method calculates the parameter in spatial mappings model, completes the demarcation to perspective projection straight line, thus realizes the demarcation of video camera.The present invention directly describes the mathematics polynomial relation of computer picture planar point and objective world point, can realize accurate camera calibration fast; This scaling method has directly reappeared the process of video camera imaging compared with the scaling method of prior art, avoids the constraint to geometric condition in video camera imaging process.

Description

A kind of camera marking method

Technical field

The present invention relates to image procossing and computer vision field, particularly relate to the corresponding scaling method of a kind of video camera shooting image and its object space position.

Background technology

One of basic task of computer vision process is the geological information that the image information obtained according to video camera calculates object in three dimensions, and rebuilds thus and identify corresponding three-dimensional body.The three-dimensional geometry position of body surface point and its mutual relationship of corresponding point in computer picture are determined by the geometric parameter model of video camera imaging, and geometric parameter is here exactly camera parameters.Under most conditions, camera parameters must just can obtain with calculating by experiment, this process is called as camera calibration, and the precision of scaling method directly has influence on the precision of computer vision process, therefore, the research that video camera carries out quick, simple and direct, accurate demarcation is significant undoubtedly.

Traditional camera marking method can divide into several classes according to calibrated reference and algorithm thinking, as: based on the camera marking method of 3D stereo target, the camera marking method based on 2D plane target drone and the camera marking method etc. based on radial constraint.These traditional camera marking methods need calibrated reference, the demarcation flow process of its classics is: first arrange calibration point, at calibration point place, fixed cameras is taken, and measures the computing machine of each calibration point as planimetric coordinates (u, v), then by each calibration point corresponding picture planimetric coordinates (u, v) and world coordinate system (X, Y, Z) substitute in camera model, according to scaling method, solve video camera measurement model parameter.

And, traditional camera marking method is all suppose that the imaging relations of video camera meets the space geometry relation of pinhole imaging system, but due to when some special request for utilizations, the design of camera lens does not meet national forest park in Xiaokeng, if still use traditional camera marking method, easily causes camera parameters to demarcate and there is theoretical error.

Summary of the invention

In order to overcome problems of the prior art, the present invention proposes a kind of camera marking method and realizes system, achieve video camera shooting image to demarcate with the corresponding of its object space position, but one as plane there being a lot of pixels, if with two points, each pixel represents that its calculated amount can very huge, therefore the design adopts space correspondent method, makes calibration process more simple and direct, convenient.

The present invention proposes a kind of camera marking method, it is characterized in that, the method comprises the following steps:

Step one, at several scale invariant feature SIFT feature points of target plane sets;

In step 2, acquisition target plane, the image coordinate of unique point, sets up unique point three dimensional space coordinate (X _w, Y _w, Z _w) and picture point coordinate (X _f, Y _f) between corresponding relation;

Step 3, this corresponding relation is substituted into spatial mappings model, this spatial mappings model is the straight line in the corresponding space of each pixel, and this straight line is determined by two points that pixel is corresponding in selected two target planes; According to selected two target plane Z=a and Z=b, gather the target image of two target planes, obtain (X in world coordinate system _w, Y _w, a) with (X _w, Y _w, b) two points, then the coordinate of these two points is substituted into proper polynomial respectively:

$\left\{\begin{array}{c}{X}_w=\underset{i=0}{\overset{n}{\Σ}}\underset{j=0}{\overset{n-i}{\Σ}}{C}_{ij}{d}_d^i{v}_d^j\\ Y_w=\underset{i=0}{\overset{n}{\Σ}}\underset{j=0}{\overset{n-i}{\Σ}}{D}_{ij}{u}_d^i{v}_d^j\end{array}\right.$

In formula, C _ijand D _ijbe respectively the polynomial expression calibrating parameters of camera model when two calibration position Zw=0mm and Zw=200mm, n is polynomial exponent number; u _d, v _dthe coordinate figure of the computing machine pixel planes point corresponding to spatial point coordinate, by what directly obtain the measurement of demarcating target, is volume coordinate point (X _w, Y _w, Z _w) corresponding to computing machine be (u as planar point _d, v _d); That i, j represent is u _d, v _dexponent number;

The C of each parameter in polynomial expression is calculated by LM algorithm _ij, D _ij, thus obtain two different polynomial expressions, utilize the coordinate of two points obtained by these two polynomial expressions, realize the demarcation of perspective projection straight line, thus complete the demarcation to video camera.

Compared with prior art, camera marking method of the present invention for model, directly describes the mathematics polynomial relation of computer picture planar point and objective world point with corresponding two the volume coordinate points of picture point, can realize accurate camera calibration fast.This scaling method has directly reappeared the process of video camera imaging compared with traditional scaling method, avoid the constraint to geometric condition in video camera imaging process, when not relying on calibrated reference, online, real-time calibration video camera internal reference, solve fast, result is accurately stable.

Accompanying drawing explanation

Fig. 1 is camera perspective projection model schematic of the present invention;

Fig. 2 is computer picture and three-dimensional body space plane mapping relations schematic diagram;

Fig. 3 is the process flow diagram of camera setting method;

Fig. 4 is that circular hole targets is marked on a map;

Fig. 5 be camera marking method of the present invention realize system schematic;

1, optical table 2, translation stage controller 3, computing machine 4, image pick-up card 5, video camera 6 (6 '), demarcation target 7 (7 '), precise electric control translation stage

Fig. 6 is the measuring error distribution curve schematic diagram of the method for camera calibration of the present invention.

61, position 1, Z _w=0mm 62, position 2, Z _w=50mm

63, position 3, Z _w=100mm 64, position 4, Z _w=150mm

65, position 5, Z _w=200mm

Embodiment

Below in conjunction with accompanying drawing, further describe specific implementation of the present invention.

In the three-dimensional body image of shot by camera, the collective entity that namely can be imaged on the spatial point on video camera pixel is a space line to the straight line in what each imaging point was corresponding is space.So corresponding to pixel each in video camera space projection straight line is demarcated, and namely obtains the position of Space Perspective Projection straight line corresponding to each pixel, thus realizes camera calibration.The method for expressing of the corresponding Space Perspective Projection straight line of pixel adopts the mode of 2 lines.As shown in Figure 1, the space that on video camera, each imaging point is corresponding is a perspective projection straight line, and in the process of demarcating, space straight line can use two, space point P ₁and P ₂show, the calibration process of whole model determines each picture point P exactly _i(such as coordinate is (u _0,v ₀).) corresponding two three dimensional space coordinate point P ₁and P ₂process.

As shown in Figure 2, in figure, the left side is space plane, and the right is image.Have the mapping relations determined between each point on image and the point on space plane, these mapping relations can be expressed as:

$\left\{\begin{array}{c}{X}_w=\underset{i=0}{\overset{n}{\Σ}}\underset{j=0}{\overset{n-i}{\Σ}}{C}_{ij}{d}_d^i{v}_d^j\\ Y_w=\underset{i=0}{\overset{n}{\Σ}}\underset{j=0}{\overset{n-i}{\Σ}}{D}_{ij}{u}_d^i{v}_d^j\end{array}\right.$

So, if set up two known plane Z in space _w=a and Z _w=b, then have the corresponding relation of image slices vegetarian refreshments and its spatial point: namely, can determine at plane Z for each pixel in image in these two known planes _w=a and Z _wexistence on=b two spatial point are corresponding with it, and the line between these two spatial point is exactly camera perspective projection straight line corresponding to this pixel, thus complete the demarcation of camera perspective projection straight line.

As shown in Figure 3, realize system for camera marking method of the present invention, achieve camera perspective projection straight line and demarcate.This system includes optical table 1, translation stage controller 2, computing machine 3, image pick-up card 4, video camera 5, demarcates target 6 (6 '), precise electric control translation stage 7 (7 ').In calibration process, demarcation target 6 is moved two positions along translation stage 7 translation direction, obtain two projection planes in these two calibration position and be respectively Z _w=a (in corresponding diagram 36,7) and Z _w=b (in corresponding diagram 36 ', 7 ').

It is below technical scheme embodiment of the present invention citing.

First, demarcation target is placed the unique point needed for several demarcation, have employed 20 calibration points altogether in the present invention's experiment, light tight sheet glass is made one group of printing opacity circular hole matrix M * N, center of circular hole position exact vertical on ranks, is uniformly distributed, and the hole heart of vertical and horizontal locality is equal and known apart from D, the circle hole radius being positioned at target center is greater than the radius of other circular holes, is called marked circle.Timing signal, target is fixed on guide rail, target is made to be positioned at whole camera field of view, marked circle is made to be positioned at the center of visual field, to mark center of circular hole for a some O, the straight line at transverse holes heart row place is x-axis, is y-axis between longitudinal hole heart place, the moving direction of target is z-axis, and camera light direction of principal axis is parallel with guide rail direction.When target moves on guide rail, determine z by the position of target plane, the world coordinates of calibration point can be determined.Then when guide rail moves to diverse location, gather corresponding target image respectively, searched for oval by Iamge Segmentation, extract the pixel coordinate of center of circular hole, so just obtain the corresponding relation of unique point and picture point).Its circular hole targets is marked on a map as shown in Figure 4.

In the process of demarcating, demarcation target plane is placed on translation stage, utilizes the distance that precise electric control translation stage 7 controls between target 6 and video camera 5; At the target position Z of these two different distance _w=a and Z _won=b, video camera 5 is utilized to take the image of target 6 (6 ') respectively;

Then, calculate image-space corresponding relation corresponding respectively on two positions respectively according to the image taking the target obtained, exponent number n adopts 5 rank usually, and its 5 rank form can be expressed as:

Xw＝C _au _d ⁰v _d ⁰+C _bu _d ⁰v _d ¹+C _du _d ⁰v _d ²+C _gu _d ⁰v _d ³+C _ku _d ⁰v _d ⁴+C _pu _d ⁰v _d ⁵+C _cu _d ¹v _d ⁰+C _eu _d ¹v _d ¹+C _hu _d ¹v _d ²+C _lu _d ¹v _d ³+C _qu _d ¹v _d ⁴+C _fu _d ²v _d ⁰+C _iu _d ²v _d ¹+C _mu _d ²v _d ²+C _ru _d ²v _d ³+C _ju _d ³v _d ⁰+C _nu _d ³v _d ¹+C _su _d ³v _d ²+C _ou _d ⁴v _d ⁰+C _tu _d ⁴v _d ¹+C _uu _d ⁵v _d ⁰；

Yw＝D _au _d ⁰v _d ⁰+D _bu _d ⁰v _d ¹+D _du _d ⁰v _d ²+D _gu _d ⁰v _d ³+D _ku _d ⁰v _d ⁴+D _pu _d ⁰v _d ⁵+D _cu _d ¹v _d ⁰+D _eu _d ¹v _d ¹+D _hu _d ¹v _d ²+D _lu _d ¹v _d ³+D _qu _d ¹v _d ⁴+D _fu _d ²v _d ⁰+D _iu _d ²v _d ¹+D _mu _d ²v _d ²+D _ru _d ²v _d ³+D _ju _d ³v _d ⁰+D _nu _d ³v _d ¹+D _su _d ³v _d ²+D _ou _d ⁴v _d ⁰+D _tu _d ⁴v _d ¹+D _uu _d ⁵v _d ⁰。

Demarcate the five rank forms adopted and altogether need demarcation 21 parameter C (Ca ~ Cu) and 21 parameter D (Da ~ Du), parameter C, D need to calculate respectively, this is measured C, D and represents polynomial parameters when Zw=0mm and Zw=200mm respectively, completes and has been demarcation to the calculating of parameter.Its computation process demarcates target point (Xw by shooting, Yw, Zw) the picture planimetric coordinates point (ud and corresponding to it, vd) be updated in this multinomial model, then by the calculating that the language program of LM least-squares algorithm has been come calibrating parameters by machine solution equation, the result of experimental calibration provides in following form.

Lavenberg-Marquardt algorithm is used to calculate parameter in spatial mappings model in the present invention, this algorithm is a kind of least square fitting algorithm, can be used for solving non-linear least square problem, be used for the places such as curve, the experimental data of each measured point is utilized the program in machine code of LM algorithm, the mensuration of above-mentioned calibrating parameters can be completed, thus complete the demarcation to perspective projection straight line.

In this example, choose Z _w=0mm and two planes corresponding to Zw=200mm two distances are as perspective projection straight line reference mark, and the calibration result obtained is as shown in table 1.

Table 1 camera parameters calibration result

In order to verify the precision of demarcation, different measuring distances use video camera take the image of target, according to the difference of displacement, on the measuring location by the measurement data of each point compared with the design data of target, analyze the difference of different measuring apart from upper actual measured results and measurement point theoretical position, thus the precision of calibrating camera is evaluated, the error distribution obtained is as shown in Figure 4.

As can be seen from measurement result, the point that removing discrete error is larger, the error of overall measurement result can control within 0.4mm.

Claims (1)

1. a camera marking method, is characterized in that, the method comprises the following steps:

Step one, at several scale invariant feature SIFT feature points of target plane sets;

In step 2, acquisition target plane, the image coordinate of unique point, sets up unique point three dimensional space coordinate (X _w, Y _w, Z _w) and picture point coordinate (X _f, Y _f) between corresponding relation;

Step 3, this corresponding relation is substituted into spatial mappings model, this spatial mappings model is the straight line in the corresponding space of each pixel, and this straight line is determined by two points that pixel is corresponding in selected two target planes; According to selected two target plane Z=a and Z=b, gather the target image of two target planes, obtain (X in world coordinate system _w, Y _w, a) with (X _w, Y _w, b) two points, then the coordinate of these two points is substituted into proper polynomial respectively:

$\left\{\begin{array}{c}{X}_w=\underset{i=0}{\overset{n}{\Σ}}\underset{j=0}{\overset{n-i}{\Σ}}{C}_{ij}{d}_d^i{v}_d^j\\ Y_w=\underset{i=0}{\overset{n}{\Σ}}\underset{j=0}{\overset{n-i}{\Σ}}{D}_{ij}{u}_d^i{v}_d^j\end{array}\right.$

In formula, C _ijand D _ijbe respectively the polynomial expression calibrating parameters of camera model when two calibration position Zw=0mm and Zw=200mm, n is polynomial exponent number; u _d, v _dthe coordinate figure of the computing machine pixel planes point corresponding to spatial point coordinate, by what directly obtain the measurement of demarcating target, is volume coordinate point (X _w, Y _w, Z _w) corresponding to computing machine be (u as planar point _d, v _d); That i, j represent is u _d, v _dexponent number;

The C of each parameter in polynomial expression is calculated by LM algorithm _ij, D _ij, thus obtain two different polynomial expressions, utilize the coordinate of two points obtained by these two polynomial expressions, realize the demarcation of perspective projection straight line, thus complete the demarcation to video camera.