WO2023156021A1

WO2023156021A1 - Assembly and method for capturing images

Info

Publication number: WO2023156021A1
Application number: PCT/EP2022/054277
Authority: WO
Inventors: Hermann Tropf
Original assignee: Vision Tools Hard- Und Software Entwicklungs-Gmbh
Priority date: 2022-02-21
Filing date: 2022-02-21
Publication date: 2023-08-24

Abstract

The assembly and the method for capturing images of a surface using two cameras in a stereo configuration simplify the process of generating the point correspondences in order to obtain a distance image. Polarized illuminations are used in a special arrangement and polarization orientation, and simple polarization filters are used with a special orientation in the respective beam path relative to the image sensors of the cameras. The process of generating point correspondences has the following advantages: the analysis of surfaces which are reflective in a diffused and specular manner both locally as well as in a proportionately mixed manner is facilitated without needing to detect the distribution of said material properties on the surface. Highlights appear at the same surface location in the two cameras. Cast shadow regions only appear in one of the two cameras and can be removed in advance from the process of determining correspondences. The image capturing process is possible with only a single shot and even with rapid movements (of the camera or the surface).

Description

Arrangement and method for taking pictures

The invention relates to an arrangement and a method for recording images of an object surface using a first camera and a second camera in a stereo arrangement. The images can be used to obtain a distance image. Distance images encode the distance of the points of the scene's surface (object points or background points) from the sensor (generally a camera) or the elevation of these points relative to a plane, in contrast to conventional images that encode gray values or colors. The pixels of a distance image therefore contain distance information (e.g. distance or height) of the associated object point that is displayed in each case. It is a classic computer vision problem.

Technical applications can be found in mechanical engineering, robotics including service robotics, electronics production, archaeology, the clothing industry, biometrics, medicine and reverse engineering.

The aim is, for example, the 3D acquisition of the fine or coarse structure or waviness or bending of a component or the localization of surface defects or an assembly control or the determination of the position of one or more objects of known geometry in a scene (example: "picking in the box "). The invention can be applied to a wide variety of surface characteristics, including mixed ones, from fully diffuse (Lambert ¹ see reflection), to matt, weakly glossy (eg paper), glossy (eg metallically glossy in various roughness levels), to partially reflective. The more shiny a surface, the narrower the reflection lobe. Knowledge of the Bidirectional Reflectance Distribution Function (BRDF) is not assumed. Fully reflective surfaces or large surface areas without a diffuse component are not considered here; deflectometry methods are suitable for this.

The methods and arrangements described here are used for triangulation with cameras in a stereo arrangement to obtain a distance image. In stereo methods, objects recorded from different camera positions are shifted and/or distorted in the images in relation to one another for systemic reasons, which ultimately contains the distance information to be found. With stereo methods, the most difficult problem is determining the correspondence of the pixels, i.e. the assignment of the pixels of one camera to the corresponding ones of the other camera. If the correspondence is known, a distance image can be calculated using triangulation if the recording geometry is known.

Triangulation is known state of the art, there is extensive, old and new literature on correspondence determination, the correspondence determination has not yet been satisfactorily solved in numerous applications. The description of this invention focuses on measures to simplify the determination of correspondence.

To determine the correspondence, features at a point PI in an image from the camera CI are compared with features at a point P2 in an image from the camera C2. The features are from data obtained by the camera at pixels PI in camera CI's image and P2 in camera C2's image (P1/P2) or there in small image sections (windows). In the course of the correspondence determination, quality values are determined for correspondence hypotheses P1/P2, which result from the comparison of the characteristics of PI with those of P2.

In order to facilitate the correspondence determination, various methods based on structured light are known. Disadvantageously, these methods require a structured light projection device.

If the transmitter and receiver are interchanged, the bidirectional reflection distribution (BRDF) is retained according to the Helmholtz reciprocity. Various approaches are based on this and work with two cameras CI and C2 and two point light sources or almost point light sources LI and L2. LI is either real or virtual in the optical axis of CI and L2 is either real or virtual in the optical axis of C2. Or LI is in close proximity to CI and L2 is in close proximity to C2 (US 7,623,701 B2, US 7,769,205 B2, US 7,574,067 B2, US 2015/0281676 A1). If an image is recorded with CI with illumination L2, and conversely an image is recorded with C2 with illumination LI, then with brilliant reflection at a surface point P the gray values of the corresponding points of P are the same.

These solutions have the disadvantage that they cannot work for Lambertian or near Lambertian reflection. With Lamberfian reflection, the luminance depends on the lighting and is independent of the viewing direction. The reason for this can be found in DE 10 2019 105 358 B4 (0037). If, in the case of Lambertian reflection, the lighting and viewing angles are interchanged at a point with unequal lighting and viewing angles, a different impression of brightness is obtained. According to DE 10 2019 105 358 B4, the lights take up a larger space or can be distributed over a larger space in a special way. In addition, an image with an illumination LX is recorded with each of the two cameras, it being possible for LX=L1 or LX=L2 to apply. Images of LI can also be taken with CI and images of L2 can also be taken with C2 (see eg 0081). Through these measures, the ability to be evaluated is achieved in a larger surface area and the ability to be evaluated is also facilitated in the case of Lambertian reflection.

In the case of point illumination as well as extended illumination according to DE 10 2019 105 358 B4, a difficulty remains when light from LI is reflected in the gloss to CI or light from L2 in the gloss is reflected to C2. The difficulty is explained below with reference to the description of FIG. A disadvantage here is that in this situation it is not possible to separate the gloss component and the diffuse component in a simple manner.

The object of the present invention is to at least partially eliminate the disadvantages mentioned.

The object is achieved according to the invention by arrangements and methods with the features of the independent patent claims. Preferred refinements and developments of the invention result from the dependent patent claims.

The invention is described in more detail using the following embodiments with associated figures. It shows:

1 shows a known arrangement with two cameras and two point light sources,

Fig. 2 shows a problem of the known arrangement with reflection in the gloss, Fig. 3a, b shows the effect of the solution according to the invention for the gloss component, 4a, b the effect of the solution according to the invention for the gloss component, with a decentralized surface point,

5 shows the solution according to the invention with reflection in the gloss,

6a, b the effect of the solution according to the invention for the diffuse component with asymmetric lighting,

7 shows the effect of the solution according to the invention for the diffuse component with symmetrical lighting,

8 an annular illumination,

9 an evaluation of the cast shadow formation, and

10 the image acquisition with a single shot, with colored surfaces.

When incident light is reflected off a surface, part of the incident light is usually specularly reflected and another part of the incident light is diffusely reflected. In order to keep the explanation of the process principles simple, a binary distinction is made in the following between diffusely reflecting and specularly reflecting reflection components, i.e. only the diffuse component or the gloss component is considered in each case. These proportions vary on the surface i.a. locally. In practice, of course, intermediate stages and locally continuous transitions are also possible.

The reflecting component depends on the angle at which the light falls and on the angle at which the camera is aimed at the surface. In the following, “in the specular angle” means that “angle of incidence = angle of reflection” applies to reflection. Depending on the width of the reflection lobe, a reflection with a gloss component also takes place at angles that deviate more or less from it. Both gray value image cameras and color image cameras can be used for the invention. In order to make the description simpler and easier to understand, the examples are initially based on greyscale cameras, and the generalization to color image cameras is given at the end.

Preliminary remarks on the figures: Lighting systems and cameras that are active in the situation described are drawn with thick lines, the others with thin lines. The drawings describe the principle and are not to scale. This also applies to the positioning of the lights; the position of the lights had to be shifted somewhat unrealistically in some cases in order to be able to accommodate various arrows so that they could be distinguished from one another.

For the sake of better understanding, the explanation in the figures uses point light sources (incandescent lamp symbols); the statements can also be made for almost point-shaped light sources and extended light sources according to DE 10 2019 105 358 B4.

Fig. 1 shows the known arrangement with two cameras CI and C2 and two point light sources LI and L2, which are located in the immediate vicinity of Cl and C2. An image of surface 1 is recorded with CI illuminated with L2 (CI, L2 drawn with thick lines). R3 is the effective reflection lobe. Correspondingly, an image with illumination LI is also recorded with C2. In the case of a brilliant reflection at a surface point P, the gray values of the corresponding points of P in CI and C2 are the same.

Fig. 2 shows the above-mentioned difficulty when light is reflected from LI in the gloss to CI ("reflection"). The Helmholtz reciprocity applies here to Ll/Cl. Here, on the one hand, an image with CI with illumination L2 and, on the other hand, an image taken with C2 with illumination LI, in CI and C2 with narrow reflection lobes the gray values of the corresponding points are both 0. This would not be particularly disturbing if it were only selectively like this, but it can certainly be over a large area on the surface 1 so be. One is therefore dependent on the diffuse component here. Since the luminance in Lambertian reflection depends on the lighting and is independent of the viewing direction, the diffuse component is the same in both cameras with the same lighting. If one takes the illumination LI for this, the diffuse component appears in C2, but in CI the diffuse component is overlaid by the strongly specular reflection from LI. On the other hand, the same applies to L2/C2.

In contrast to the prior art, the invention presented here uses polarized illumination in a special arrangement and polarization direction and assignment to the cameras, as well as simple polarization filters with a special polarization direction in the respective beam path to the image recorders (image sensors) of the cameras. Simple linear filters are sufficient as polarization filters; alternatively, circular filters can also be used (with a suitable sequence of linear filters/Lambda quarter plate).

To create a distance image of a surface 1 using two cameras CI and C2 in a stereo arrangement and with two lights, the two lights LI and L2 are arranged so that when rays from the first light LI are reflected on a surface point P in the glancing angle to the second camera C2 , rays from the second illumination L2 are also reflected at least approximately at the same surface point P in the glancing angle to the first camera CI.

The arrangement is characterized in that there is a polarization filter in the beam path of each camera, each with a polarization direction, the polarization direction of the first camera rC1, and the polarization direction of the second camera rC2, and that the illumination LI operates with a first polarization direction rLl and the second further illumination L2 operates with a second polarization direction rL2, and that the camera and illumination polarization directions are oriented such that light from LI reflecting on the surface toward camera CI is substantially blocked at Rcl and that light from L2 specularly reflected at the surface towards camera C2 is substantially blocked at Rc2. and that light specularly reflected from LI to C2 is substantially transmitted at Rc2 and light specularly reflected from L2 to CI is substantially transmitted at Rcl.

The polarization filters can be located in the beam path of the cameras in front of the lens or in the lens or between the lens and the image recorder of the respective camera. The polarization filters according to the invention in the beam path of the cameras are also implemented in polarization cameras with their (typically 2×2) pixel polarization filter arrangements of different polarization directions, in which the pixels of a single, selected polarization direction can be detected.

The polarization directions are set up and fixed when configuring the system or in the event of a service case according to the conditions described and are retained for different surfaces during operation. For the last condition (LI according to C2 and L2 according to CI) one advantageously uses a central point of the entire evaluation area and an average surface inclination to be expected there.

Explanation of the gloss content Figures 3a and 3b show the solution for the gloss component, with the reflection lobes R3. Camera CI and lighting L2 are active in FIG. 3a, and camera C2 and lighting LI are active in FIG. 3b. A central point P of the surface is considered in FIG. 3a/b. In this case, the gloss components are the same in both cameras, since this light component, reflected from LI to C2, is essentially transmitted and the light component, reflected from L2 to CI, is essentially transmitted.

If a surface point is decentralized, as in FIGS. 4a and 4b (backwards or forwards from the direction of the observer in FIGS. 4a/b), the direction of polarization is rotated by an angle with specular reflection from LI to C2 and from L2 to CI rotated by the same angle, only with the opposite sign (seen from the direction of the viewing camera). Due to the filters rCl/rC2, this results in a weakening of the gloss component, but this weakening is the same for both cameras, since the angles of rotation are the same in terms of absolute value.

Regarding the directions of polarization drawn in the figures: The directions of polarization of the corresponding polarization filters are drawn in both for the illuminations and for the cameras; for the sake of clarity, the figures simply distinguish between HÖR, horizontally hatched, and VER, vertically hatched. This corresponds to the following special case: with VER the polarization direction is at least approximately in the plane of incidence for a central point, with HÖR transverse to it. Here, the plane of incidence is spanned by the incident beam and the perpendicular to a surface element. The same applies to the polarization direction arrows R3 drawn horizontally/vertically. In Fig. 4a/b, the symbolic directional arrows 3 are drawn rotated with a rotated direction of polarization. The only important thing is that the polarized components of LI are essentially blocked by the filters in CI, as are the polarized components of L2 in C2. Conversely, in CI, the polarized components of L2 are essentially transmitted, as are the polarized components of LI in C2.

So you can rotate all HOR filters as well as all VER filters by e.g. 45 degrees and achieve the same desired effects.

You won't always get full transmission or blocking, but the effects are sufficient to get the desired results.

Fig. 5 shows the solution to the above problem when light is reflected from LI in the specular to CI. The filter rCl prevents the polarized light from reaching CI from LI.

When LI and CI are close together, the reflection of FIG. 5 corresponds to a specular reflection from a plane perpendicular to the polarized illumination beam. In the case of specular back-reflection, there is indeed a phase jump, but the direction of polarization is retained and thus the blocking by the polarization filter rCl, which is transverse to the direction of polarization. This effect is weakened only slightly if the LI/CI are further apart.

Using polarized lighting in order to then partially remove the polarized component from the evaluation in the cameras seems absurd at first glance. The advantages become clear when you look at this in the context of the diffuse component.

Explanation of the diffuse portion

In general, when polarized light hits a Lambertian diffuse reflecting surface, it becomes depolarized. When LI is illuminated, part of the diffusely reflected, ie depolarized component through the polarization filter rCl. The same applies to L2 and rC2.

6a shows the diffuse component R2 when illuminated by LI. As mentioned above, in the case of diffuse reflection, the luminance is independent of the viewing direction and is therefore in principle the same for both cameras. The diffuse component is indicated in the figures by the circle R2. The star symbol 2 is intended to indicate unpolarized light. The arrow symbol 3 is intended to indicate polarized light. The two polarization filters rC1 and rC2 let through the same portion of the diffuse portion that is the same for both cameras. This applies to any tilt angle of the surface 1 at the reflecting point.

6b shows the situation when illuminated by L2. Here the illumination angle is flatter and the diffuse light component is correspondingly weaker, indicated by the smaller circle R2 in FIG. 6b.

FIG. 7 shows the diffuse component for a symmetrical situation, that is to say a symmetrical lighting/camera position. It should be noted that in a symmetrical situation, when the specular parts L1/C2 and L2/C1 are the same (Figures 3 and 4), with locally isotropic surfaces, the diffuse parts Ll/Cl, L1/C2, L2/C1, L2/C2 are all are the same (in terms of amount, the same angle of incidence).

The first illumination LI is advantageously located in a ring around the optical axis of the first camera CI, see FIG. 8, and the second illumination L2 is in a ring around the optical axis of the second camera C2. The conditions for using the Helmholtz reciprocity (illumination Ll/2 on the optical axis of Cl/2) are thus well approximated.

To determine the correspondences: For the sake of clarity, we assume that 4 separate images are taken:

With camera CI (“left”) under illumination of LI (“left”), Livonli for short.

With camera CI (“left”) under illumination of L2 (“right”), Livonre for short.

With camera C2 (“right”) under illumination of LI (“left”), Revonli for short.

With camera C2 (“right”) under illumination of L2 (“right”), Revonre for short.

For each of these images there is a gloss component and a diffuse component, in short livonli gloss, livonli diffuse, livonre gloss, etc

Livonli = Livonli shine + Livonlidiffuse,

Livonre = Livonre Luster + Livonrediffuse, etc.

Due to Livoni gloss=0, Revonre gloss=0 (blocking by crossed polarization directions rL1 and rCl, as well as rL2 and rC2) applies

Livonre + Livonli

= Livonre gloss + Livonrediffuse + Livonlidiffuse, and analogously:

Revonli + Revonre

= revonli gloss + revonlidiffuse + revonrediffuse.

With that applies

Livonre + Livonli = Revonli + Revonre because of Revonli gloss = Livonre gloss and

Revonlidiffus=Livonlidiffus (diffuse, same lighting, i.e. independent of viewing direction), and

Livonrediffus=Revonrediffus (diffuse, same lighting, i.e. independent of viewing direction) .

These two images (Livonre+Livonli) and (Revonli+Revonre) are advantageously used to determine correspondence, which can be done using known methods.

It is simply sufficient to leave both lights LI and L2 switched on in order to record the sum of Livonre + Livonli in CI and at the same time the sum of Revonli + Revonre in C2. It is thus possible to take a picture with a single shot.

However, if use is to be made of the method for evaluating cast shadows explained in DE 10 2019 105 358 B4, description of FIG. 23, a separate evaluation of Livonre and Revonli first appears necessary.

For single shot image acquisition:

FIG. 9 shows the evaluation of the cast shadow formation according to DE 10 2019 105 358 B4, FIG. 23; L2 is switched on to determine the cast shadow areas 4 visible in camera CI, which are not visible in camera C2 and therefore cannot contribute to the determination of correspondence. Symmetrical to this, the cast shadow areas visible in camera C2, which are not visible in camera CI and therefore not for correspondence determination, are determined can contribute, LI turned on. This is done by evaluating Livonre and Revonli at separate times. With this temporal separation, 2 shots are therefore required.

To evaluate the formation of hard shadows, the image can be recorded with a single shot on achromatic surfaces by separating them by color. For this purpose, the lights LI and L2 can be operated with different colors F1 or F2, and the two cameras are color cameras that are able to separate these two colors.

Image acquisition with a single shot is also possible with colored surfaces as follows, see Fig. 10:

The illumination LI takes place with two spectral ranges, sLO and SL1. The lighting L2 is done with two spectral ranges, sLO and sL2. After the image recording, the camera CI can work in a spectral range sCO that overlaps with spectral range sLO and in a spectral range sC2 that overlaps with spectral range sL2. After the image recording, the camera C2 can work in the spectral range sCO and in a spectral range sCl, which overlaps with the spectral range sL1. The spectral ranges sL1 and sCO do not overlap, or only slightly, as do the spectral ranges sL2 and sCO.

And the spectral ranges sL1 and sC2 do not or only slightly overlap, as do the spectral ranges sL2 and sCl.

Technically, lighting spectral ranges sL0, sL1 and sL2 can preferably be realized by assembling narrow-band light-emitting diodes, as indicated by arrows in FIG. 10; With conventional color cameras, the spectral ranges sCO, sC1 and sC2 are broadband ranges, as indicated by curves in FIG. The images recorded by the cameras C1 and C2 can be further processed to determine the cast shadow. For this purpose, camera CI processes the image it has recorded in the spectral range sC2, and camera C2 processes the image it has recorded in the spectral range sCl. Under the circumstances described, the drop shadow areas appear black or nearly black in the images captured by the two cameras. This makes it easy to determine the shadow areas.

Here it must be ensured that the parts of the surface that are visible in both cameras still deliver, albeit small, parts that can still be evaluated, otherwise these parts of the surface also appear black and are therefore indistinguishable from the hard shadow areas. Depending on the surface color, a subtractive color mixture can take place on the surface parts visible in both cameras, so that they also appear black. Fortunately, the phenomenon that the gloss component is reflected without color mixing, and the gloss component is let through by the polarization filters in the operating mode relevant here (light from LI to C2 and from L2 to CI) is fortunate.

When using multispectral cameras, the requirements for the spectral ranges can be met very well, with the specified spectral ranges not having to be contiguous, i.e. they can consist of disjunctive individual ranges.

General Benefits

According to the invention, distance images of a surface are made possible by means of stereo analysis without movement (the laser light section requires movement that must be synchronized with the signal processing), without complex pattern projection devices, with lighting units that are easy to implement and with inexpensive, easily available on the market available standard greyscale or color cameras. It is with the same arrangement the evaluation of both locally as well as proportionately mixed diffuse and specular reflecting surfaces possible without knowing how these material properties are distributed on the surface.

In time-of-flight systems, for physical reasons, accuracy limits are set regardless of the image field size and the camera resolution, even for small image fields; in contrast to this, the achievable accuracy can in principle be increased at will with increasing camera resolution and/or by "stitching" several small image fields.

Highlights, otherwise very disturbing, since with conventional stereo in the two cameras i.a. appearing at different surface locations, appearing at the same surface location here. An otherwise serious problem with shiny spots becomes an advantage.

In conventional stereo, drop shadows are sometimes visible in both cameras, sometimes only in one, and if visible in both cameras, they generally appear with different brightness. And they are difficult to distinguish from dark surface regions visible in both cameras. Cast shadows therefore represent a major problem with conventional stereo. With the arrangement according to the invention, cast shadow areas only appear in one of the two cameras and can be removed from the correspondence determination in advance.

It is possible to take a picture with a single shot and can therefore also take place with fast movement (camera or surface).

Further variants of the invention

Color: Color cameras can be used to advantage for colored surfaces: Instead of comparing individual gray values or contents of gray value windows when determining correspondence, you only need to compare the color values or contents of color value windows, eg the red-green-blue values. In the case of colored surfaces, this simplifies the evaluation compared to pure intensity evaluation and improves the result.

Special camera types:

Of course, HDR cameras can also be used to expand the dynamic range.

The invention is not limited to the use of 2D cameras, but line cameras, ie cameras in which only one or very few directly or closely adjacent lines are scanned, can also be used without modifications.

Multiple Camera Pairs:

The previous considerations relate to only one single stereo camera pair. With several pairs of cameras, a separate distance image can be created for each camera pair; several distance images can be merged using known methods to form a common distance image, which increases the accuracy and/or enlarges the evaluation area ("stitching").

Evaluation unit:

The image evaluation of the spectral components can be carried out in at least one of the two cameras CI, C2 ("intelligent cameras") or in one of the cameras mera's separate evaluation unit such as a PC or in an evaluation unit to which only the image sensors of the cameras are connected. In this sense, “the camera C. . . . can work in a spectral range after the image has been recorded" and "the camera C... processes the image it has recorded".

Claims

Expectations

1. Arrangement for recording images of a surface (1) by means of a first camera (CI) for recording at least a first image of the surface (1) and a second camera (C2) for recording at least a second image of the surface (1) in a stereo arrangement and having a first illumination (LI) and a second illumination (L2), the first illumination (LI) and the second illumination (L2) being arranged such that when rays from the first illumination (LI) are incident on a surface point ( P) are reflected in the glancing angle to the second camera (C2), rays from the second illumination (L2) are reflected at least approximately at the same surface point (P) in the glancing angle to the first camera (CI), characterized in that in the beam path of the first camera (CI) a first polarization filter with a first polarization direction (rCl) is positioned and a second polarization filter with a second polarization direction (rC2) is positioned in the beam path of the second camera (C2), and that the first illumination (LI) with a first polarization direction ( rLl) can work and the second illumination (L2) can work with a second polarization direction (rL2), and that the camera and illumination polarization directions are oriented such that light from the first illumination (LI) specularly reflected at the surface downstream of the first camera (CI) is substantially blocked by the first polarization filter (rCl), and that light from the second illumination (L2), which is specularly reflected at the surface after the second camera (C2), is substantially blocked by the second polarizing filter (rC2), and that light which comes from the first illumination (LI) after the second camera (C2) is specularly reflected is essentially transmitted by the second polarizing filter (rC2), and light which is specularly reflected by the second illumination (L2) after the first camera (CI) passes through the first polarizing filter im essentially is passed.

2. Arrangement according to claim 1, wherein the first illumination (LI) is arranged annularly around the optical axis of the first camera (CI) and the second illumination (L2) is arranged annularly around the optical axis of the second camera (C2).

3. Arrangement according to one of the preceding claims, characterized in that the first lighting (LI) with a first color (F1) and the second lighting (L2) with a second color (F2) different from the first color (F1) can be operated and the two cameras (C1, C2) are color cameras capable of separating the two colors (F1 and F2).

4. Arrangement according to claim 1 or 2, characterized in that

- the first illumination (LI) can be operated with a general illumination spectral range (sLO) and a first special illumination spectral range (sLl),

- the second illumination (L2) can be operated with the general illumination spectral range (sLO) and a second special illumination spectral range (sL2), - An evaluation of at least one by the first camera (CI) recorded first image in a general camera spectral range (sCO) that overlaps with the general illumination spectral range (sLO), and in a second camera spectral range (sC2) with the second illumination spectral range ( sL2) overlaps, can take place and

- At least one second image recorded by the second camera (C2) can be evaluated in the general camera spectral range (sCO) and in a first camera spectral range (sCl) which overlaps with the first illumination spectral range (sLl),

- wherein the first illumination spectral range (sL1) and the general camera spectral range (sCO) do not or only slightly overlap each other, and wherein the second illumination spectral range (sL2) and the general camera spectral range (sCO) do not or only slightly overlap each other,

- And wherein the first illumination spectral range (sLl) and the second camera spectral range (sC2) do not or only slightly overlap each other, and the second illumination spectral range (sL2) and the first camera spectral range (sCl) do not or only slightly overlap each other.

5. Arrangement according to one of the preceding claims, further comprising a device for obtaining corresponding pairs of points with respect to points on the object surface (1) from the at least one first image and the at least one second image.

6. Arrangement according to one of the preceding claims, further comprising a device for obtaining a distance image of the object surface (1) from the at least one first image and the at least one second image.

7. Method for recording images of a surface (1) using a first camera (CI) and a second camera (C2) in a stereo arrangement and with a first illumination (LI) and a second illumination (L2), the first illumination (LI ) and the second illumination (L2) are arranged such that when rays from the first illumination (LI) are reflected on a surface point (P) in the specular angle to the second camera (C2), rays from the second illumination (L2) on at least are reflected approximately at the same surface point (P) in the glancing angle to the first camera (CI), characterized in that a first polarization filter with a first polarization direction (rCl) is positioned in the beam path of the first camera (CI) and in the beam path of the second camera (C2 ) a second polarization filter with a second polarization direction (rC2) is positioned, and that the first illumination (LI) operates with a first polarization direction (rLl) and the second illumination (L2) operates with a second polarization direction (rL2), and that the camera - and illumination polarization directions are oriented such that light from the first illumination (LI) specularly reflected at the surface downstream of the first camera (CI) is substantially blocked by the first polarization filter (rCl), and that light from the second illumination (L2) specularly reflected at the surface after the second camera (C2) is substantially blocked by the second polarizing filter (rC2), and that light emitted by the first illumination (LI) after the second camera (C2) is specularly reflected, is essentially transmitted by the second polarization filter (rC2), and light that is specularly reflected by the second illumination (L2) after the first camera (CI) is essentially transmitted at the first polarization filter, and in that at least a first image of the surface (1) is recorded with the first camera (CI) and at least a second image of the surface (1) is recorded with the second camera (C2).

8. The method according to claim 7, wherein the first illumination (LI) is arranged annularly around the optical axis of the first camera (CI) and the second illumination (L2) is arranged annularly around the optical axis of the second camera (C2).

9. The method as claimed in claim 7 or 8, characterized in that the first lighting (LI) is operated with a first color (F1) and the second lighting (L2) is operated with a second color (F2) which is different from the first color (F1). and the two cameras (C1, C2) are color cameras capable of separating the two colors (F1 and F2).

10. The method according to claim 7 or 8, characterized in that

- the first illumination (LI) is operated with a general illumination spectral range (sLO) and a first special illumination spectral range (sLl),

- the second illumination (L2) is operated with the general illumination spectral range (sLO) and a second special illumination spectral range (sL2),

- the at least one first image recorded by the first camera (CI) is fed to an evaluation in a general camera spectral range (sCO) which overlaps with the general illumination spectral range (sLO), and in a second camera spectral range (sC2) which overlaps with the second Illumination spectral range (sL2) overlapped, - the at least one second image recorded by the second camera (C2) is fed to the evaluation in the general camera spectral range (sCO) and in a first camera spectral range (sCl), which overlaps with the first illumination spectral range (sLl),

11. The method according to any one of claims 7 to 10, characterized in that the at least one first image and the at least one second image are used to obtain corresponding pairs of points with respect to points on the object surface (1).

12. The method as claimed in one of claims 7 to 11, characterized in that the at least one first image and the at least one second image are used to obtain a distance image of the object surface (1).