CN101858741A - Zoom ranging method based on single camera - Google Patents

Abstract

The invention belongs to the technical field of the computer stereo vision, relates to a zoom ranging method based on a single camera, and provides a novel stereo vision model which can be used for acquiring depth information of a two-dimensional image. In the method, three-dimensional coordinates of characteristic points of a scene in a camera coordinate system can be obtained through image processing and analysis only after one digital camera with an optical zoom respectively images the scene once under two different focal lengths. The concept of the stereo vision model provided by the invention is novel, thereby opening up a new direction for the theoretical study of the computer stereo vision. Compared with the currently popular binocular stereo vision and other vision ranging techniques, the stereo vision model can be used more easily, and the broad application prospects thereof can be predicted.

Description

A kind of zoom distance-finding method based on one camera

Technical field

The present invention relates to the computer stereo vision field, disclose a kind of zoom distance-finding method specifically, proposed a kind of new stereoscopic vision model, can obtain depth information from two dimensional image by this model based on one camera.

Background technology

Obtaining each point in the scene is one of vital task of computer vision system with respect to the distance of video camera.The vision ranging technology of comparative maturity is mainly based on following several models at present: binocular stereo vision, structured light method, geometrical optics approach.Certainly also has the application model under a lot of specific environments, here no longer narration.

Wherein of paramount importance is exactly the binocular tri-dimensional vision model, and this model is made of two identical video cameras, and two video cameras spatially exist rotation or translation relation.In this model, the image space difference of same unique point on two camera review planes in the scene, it is right that these two picture points are called match point.We claim that two distances between the video camera projection centre are baseline, during two width of cloth doublings of the image match point between distance be parallax.Calculate by Stereo Matching Technology in the Flame Image Process and parallax, just can obtain the depth information of object point in the scene.

Structured light range finding imaging system then uses principle of triangulation to come compute depth.In a simple spot projection system, at a distance of a parallax range,, just can obtain the depth information of scene between projection light source instrument and the video camera by determining parameters such as image space behind the scene point reflection source light and projected angle.The general laser that adopts is as secondary light source.

Geometrical optics approach mainly comprises focusing and out of focus method.The ultimate principle of focusing is by adjusting the image distance of video camera, make imaging plane be in focal position with respect to the measured point, and under the known condition of focal length and image distance, but the scioptics imaging formula is tried to achieve object distance.The focusing principle is fairly simple, but its precision is subjected to the serious restriction of hardware, and different depth zone needs to focus on again, thus measure loaded down with trivial details slowly, so use seldom.The out of focus method has avoided seeking the operation of vernier focusing position, the size of imaging circle spot when it utilizes the object point out-focus, and promptly the fog-level of image obtains depth information.But the accurate demarcation of out of focus model is its difficulty place.

Summary of the invention

The object of the invention is to propose a kind of new stereoscopic vision model, can obtain the depth information of two dimensional image by this model.Its value is that this model opened up new direction for the computer stereo vision Study on Theory.

The zoom distance-finding method that the present invention proposes based on one camera, only need with a digital camera that can be optical-zoomed under two different focal lengths to Same Scene respectively imaging once just can draw the three-dimensional coordinate of unique point in camera coordinates system in the scene by Flame Image Process and analytical technology.Its operating process mainly contains following a few step:

Step 1: camera is aimed at the scene that will find range, and fixing;

Step 2: in focal distance f ₁, f ₂Respectively imaging once and is preserved the image of being gathered for twice as a stereoscopic image, generally claims the prospect map that is imaged as under less focal length, than being imaged as close shot figure under the focal length greatly; (be without loss of generality, can suppose focal distance f ₁＜f ₂)

Step 3: the coupling of finishing stereo-picture;

Step 4: the three-dimensional coordinate of calculated characteristics point on request.

The solid coupling that a wherein the most key step is an image, noticeable place have following some:

One, can adopt the SIFT characteristic matching to calibrate the public scene of two width of cloth images in prospect map, because prospect map is compared with close shot figure, the spatial dimension that prospect map comprises is wider, and close shot figure is exactly the public scene of two width of cloth images;

Two, selected pixels point in the public scene of prospect map earlier, and then in the close shot figure its corresponding match point of search.Because the details of scene is described more horn of plenty among the close shot figure, it should comprise all details of public scene in the prospect map, and promptly the picture point of prospect map all has match point in the public scene of two width of cloth images in close shot figure, and not necessarily sets up conversely;

Three, can be in subrange the search matched point, because under the ideal model, is in the image coordinate system set up of initial point with the photocentre, two corresponding mutually match points are crossed the slope of photocentre should be identical, and among the close shot figure polar radius of match point greater than the polar radius of corresponding point in the prospect map.

Description of drawings

Fig. 1 is the zoom range measurement principle figure based on one camera of the present invention.

Fig. 2 is the space plane cut-away view of zoom range finding model of the present invention.

Fig. 3 is a stereoscopic image of the present invention.

Embodiment

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

The zoom range measurement principle figure based on one camera as shown in Figure 1, the same pin-hole imaging model that also is based on the binocular tri-dimensional vision model of this model carries out geometric analysis.With photocentre o is initial point, sets up the rectangular coordinate system in space oxyz of camera, and its midplane xoy is the plane of delineation.Fixedly camera to the same space respectively in focal distance f ₁, f ₂Each imaging once (is supposed f here down ₁＜f ₂), any 1 s equals f at focal length in the scene ₁The time be imaged as s ₁Equal f at focal length ₂The time be imaged as s ₂The model of twice imaging is drawn in the same coordinate system analyzes.Picture point s under the ideal model ₁, s ₂All on straight line om, s s ₁Be focal length f ₁Following lens center o ₁Straight line, s s ₂Be focal length f ₂Following lens center o ₂Straight line, o o wherein ₁=f ₁, o o ₂=f ₂Plane som was the plane of z axle, and perpendicular to picture plane xoy.

Plane taken som separate analysis, as shown in Figure 2.r ₁, r ₂Be picture point s ₁, s ₂To the distance of photocentre, i.e. polar radius in the xoy of plane, as shown in Figure 3.Here suppose f ₁＜f ₂, abcd represents the size as the plane among Fig. 3, a ' b ' c ' d ' is the public scene of presentation video 1 and image 2 then.Because f ₁＜f ₂So image 1 is compared with image 2, image 1 comprises wider space, and comprises more details as 2.We can be earlier get 1 s arbitrarily in that a ' b ' of image 1 c ' d ' is interior ₁, by finding s on relevant (as value of color) the straight line om in image 2 of feature ₁Match point s ₂So just can determine r ₁, r ₂Value.Therefore ∠ α, ∠ β can separate triangle s o among Fig. 2 ₁o ₂Also can separate.

As shown in Figure 2, line segment os puts the distance of s to photocentre, i.e. polar radius r in camera coordinate system oxyz in the scene.By above analysis triangle soo as can be known ₁Also can separate.The coordinate of s in camera coordinate system oxyz of setting up an office is for (x, y z), then have x=rsin γ cos θ, y=rsin γ sin θ, z=rcos γ.

Only require so to solve r, sin γ, cos γ, sin θ, cos θ, just can determine the coordinate of a s in camera coordinates is xyz (z), its solution procedure is as follows for x, y:

One, establishes some s in the xoy of plane ₁Coordinate be (i, j), match point s ₂Coordinate be (k n) then has:

$\sin \mathrm{\θ}=-\frac{i}{\sqrt{i^2+j^2}}---\left(1\right)$

$\cos \mathrm{\θ}=-\frac{j}{\sqrt{i^2+j^2}}---\left(2\right)$

${\mathrm{<figure-callout\; id="1"\; label="r"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">r</figure-callout>}}_1=\sqrt{{\mathrm{<figure-callout\; id="2"\; label="i"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">i</figure-callout>}}^2+j^2}---\left(3\right)$

${\mathrm{<figure-callout\; id="2"\; label="r"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">r</figure-callout>}}_2=\sqrt{{\mathrm{<figure-callout\; id="2"\; label="k"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">k</figure-callout>}}^2+n^2}---\left(4\right)$

Two, find the solution triangle s o earlier ₁o ₂:

If d=o ₁o ₂=f ₁-f ₂, a=s o ₁, b=s o ₂

By above derivation as can be known:

$\sin \mathrm{\&alpha;}=\frac{r_1}{\sqrt{{\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">f</figure-callout>}}_1^2+{\mathrm{<figure-callout\; id="1"\; label="r"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">r</figure-callout>}}_1^2}}---\left(5\right)$

$\cos \mathrm{\&alpha;}=\frac{f_1}{\sqrt{{\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">f</figure-callout>}}_1^2+{\mathrm{<figure-callout\; id="1"\; label="r"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">r</figure-callout>}}_1^2}}---\left(6\right)$

$\tan \mathrm{\&alpha;}=\frac{r_1}{f_1}---\left(7\right)$

$\cot \mathrm{\&alpha;}=\frac{f_1}{r_1}---\left(8\right)$

$\sin \mathrm{\&beta;}=\frac{r_2}{\sqrt{{\mathrm{<figure-callout\; id="2"\; label="f"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">f</figure-callout>}}_2^2+{\mathrm{<figure-callout\; id="2"\; label="r"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">r</figure-callout>}}_2^2}}---\left(9\right)$

$\cos \mathrm{\&beta;}=\frac{f_2}{\sqrt{{\mathrm{<figure-callout\; id="2"\; label="f"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">f</figure-callout>}}_2^2+{\mathrm{<figure-callout\; id="2"\; label="r"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">r</figure-callout>}}_2^2}}---\left(10\right)$

$\tan \mathrm{\&beta;}=\frac{r_2}{f_2}---\left(11\right)$

$\cot \mathrm{\&beta;}=\frac{f_2}{r_2}---\left(12\right)$

Have

d ²＝a ²+b ²-2ab?cos(β-α) (13)

a ²＝d ²+b ²-2db?cos(180-β) (14)

b ²＝d ²+a ²-2da?cosα (15)

With (15) substitution (13), (14) substitution (15)

a＝b?cos(β-α)+d?cosα (16)

d＝b?cos(180-β)+a?cosα (17)

Again (16) substitution (17) is got

$b=\frac{d\sin \mathrm{\&alpha;}}{\sin \mathrm{\&alpha;}\cos \mathrm{\&beta;}+\cos \mathrm{\&alpha;}\sin \mathrm{\&beta;}}$

$=\frac{d}{\tan \mathrm{\&alpha;}\cos \mathrm{\&beta;}+\cot \mathrm{\&alpha;}\sin \mathrm{\&beta;}}---\left(18\right)$

With (18) substitution (16)

$a=\frac{\sin \mathrm{\&beta;}+2\sin \mathrm{\&alpha;}\cos \mathrm{\&alpha;}\cos \mathrm{\&beta;}}{\sin \mathrm{\&alpha;}\cos \mathrm{\&beta;}+\cos \mathrm{\&alpha;}\sin \mathrm{\&beta;}}d$

$=\frac{\tan \mathrm{\&beta;}/\sin \mathrm{\&alpha;}+2\cos \mathrm{\&alpha;}}{1+\cot \mathrm{\&alpha;}\tan \mathrm{\&beta;}}d---\left(19\right)$

Three, find the solution triangle soo again ₁:

$r=\sqrt{{\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">f</figure-callout>}}_1^2+a^2+2f_{1}a\cos \mathrm{\&alpha;}}---\left(20\right)$

$\cos \mathrm{\&gamma;}=\frac{{\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">f</figure-callout>}}_1^2+r^2-{\mathrm{<figure-callout\; id="2"\; label="a"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">a</figure-callout>}}^2}{2f_{1}r}$

$=\frac{f_1+a\cos \mathrm{\&alpha;}}{r}---\left(21\right)$

$\sin \mathrm{\&gamma;}=\sqrt{1-{\mathrm{<figure-callout\; id="2"\; label="cos"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">cos</figure-callout>}}^2\mathrm{\&gamma;}}$

$=\frac{a\sin \mathrm{\&alpha;}}{r}---\left(22\right)$

So just finished finding the solution of r, sin γ, cos γ, sin θ, cos θ, by formula

$\left\{\begin{array}{cc}x=r\sin \mathrm{\&gamma;}\cos \mathrm{\&theta;}& \left(23\right)\\ y=r\sin \mathrm{\&gamma;}\sin \mathrm{\&theta;}& \left(24\right)\\ z=r\cos \mathrm{\&gamma;}& \left(25\right)\end{array}\right.$

But the coordinate of invocation point s in camera coordinates is oxyz is

$\left\{\begin{array}{cc}x=j(f_1-f_2)\frac{\sqrt{{\mathrm{<figure-callout\; id="2"\; label="k"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">k</figure-callout>}}^2+{\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">n</figure-callout>}}^2}+2f_1{f}_2[1-{\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">f</figure-callout>}}_1^2/({\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">f</figure-callout>}}_1^2+{\mathrm{<figure-callout\; id="2"\; label="i"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">i</figure-callout>}}^2+j^2)]/\sqrt{{\mathrm{<figure-callout\; id="2"\; label="i"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">i</figure-callout>}}^2+j^2}}{f_2\sqrt{{\mathrm{<figure-callout\; id="2"\; label="i"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">i</figure-callout>}}^2+{\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">j</figure-callout>}}^2}+f_1\sqrt{{\mathrm{<figure-callout\; id="2"\; label="k"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">k</figure-callout>}}^2+n^2}}& \left(26\right)\\ y=i(f_1-f_2)\frac{\sqrt{{\mathrm{<figure-callout\; id="2"\; label="k"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">k</figure-callout>}}^2+{\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">n</figure-callout>}}^2}+2f_1{f}_2[1-{\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">f</figure-callout>}}_1^2/({\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">f</figure-callout>}}_1^2+{\mathrm{<figure-callout\; id="2"\; label="i"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">i</figure-callout>}}^2+j^2)]/\sqrt{{\mathrm{<figure-callout\; id="2"\; label="i"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">i</figure-callout>}}^2+j^2}}{f_2\sqrt{{\mathrm{<figure-callout\; id="2"\; label="i"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">i</figure-callout>}}^2+{\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">j</figure-callout>}}^2}+f_1\sqrt{{\mathrm{<figure-callout\; id="2"\; label="k"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">k</figure-callout>}}^2+n^2}}& \left(27\right)\\ z={\mathrm{<figure-callout\; id="2"\; label="f"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">f</figure-callout>}}_2+(f_2-f_1)\frac{{2\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">f</figure-callout>}}_1^2/({\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">f</figure-callout>}}_1^2+{\mathrm{<figure-callout\; id="2"\; label="i"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">i</figure-callout>}}^2+j^2)-f_2\sqrt{{\mathrm{<figure-callout\; id="2"\; label="i"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">i</figure-callout>}}^2+j^2}}{f_2(\sqrt{{\mathrm{<figure-callout\; id="2"\; label="i"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">i</figure-callout>}}^2+{\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">j</figure-callout>}}^2}+f_1\sqrt{{\mathrm{<figure-callout\; id="2"\; label="k"\; filenames="HSA00000115853400011.png"\; state="\{\{state\}\}">k</figure-callout>}}^2+n^2})}& \left(28\right)\end{array}\right.$

Wherein (i, j), (k, n) match point is to s ₁, s ₂Coordinate, f ₁, f ₂Be focal length, and f ₁＜f ₂Because be the handstand imaging, so the coordinate points in the xoy of plane of spatial point s and imaging point are always in opposite interval.If world coordinate system is overlapped with camera coordinates system, the three-dimensional coordinate that then records is absolute coordinates under the world coordinate system.

Claims (1)

1. zoom distance-finding method based on unit, it is characterized in that: need to move a segment distance along optical axis direction with a digital camera, and two positions before and after moving to Same Scene respectively imaging once just can draw the three-dimensional coordinate of unique point in camera coordinates is in the scene by Flame Image Process and analytical technology;

Its operating process may further comprise the steps:

Step 1: camera and slide mechanism thereof are aimed at the scene that will find range;

Step 2: utilize the slip structure that camera is moved a segment distance along optical axis direction, and two positions before and after moving to the imaging of Same Scene difference once, the image of being gathered for twice is preserved as a stereoscopic image, suppose that camera moves forward, general be imaged as prospect map before claiming to move, before and after moving, be imaged as close shot figure;

Step 3: the coupling of finishing stereo-picture: adopt the SIFT characteristic matching in prospect map, to calibrate the public scene of two width of cloth images because prospect map and close shot figure mutually this, the spatial dimension that prospect map comprises is wider, and close shot figure is exactly the public scene of two width of cloth images; Selected pixels point in the public scene of prospect map then, and then in close shot figure, search for its corresponding match point, because the details of scene is described more horn of plenty among the close shot figure, it should comprise all details of public scene in the prospect map, promptly the picture point of prospect map all has match point in the public scene of two width of cloth images in close shot figure, and not necessarily sets up conversely; At last can be in subrange the search matched point, because under the ideal model, is in the image coordinate system set up of initial point with the photocentre, two corresponding mutually match points are crossed the slope of photocentre should be identical, and among the close shot figure polar radius of match point greater than the polar radius of corresponding point in the prospect map.

Images