DE102022123295A1 - Stereo vision system, driver assistance system and method for generating a stereo vision from two 2D representations - Google Patents

Abstract

The invention relates to a stereo vision system which is designed to create a 3D representation (16) from 2D representations (7, 8) of two monocular imaging units (2, 3) observing a scene (6) by deriving depth data (TPS). To reconstruct scene (6), at least one monocular imaging unit (3) being sensitive to radiation outside a visible spectrum.

Description

The invention relates to a stereo vision system which is designed to output a representation of the scene from the superimposed signals from signals from at least two monocular imaging units observing a scene, a driver assistance system comprising the stereo vision system, and a method for generating a stereo vision from two 2D -Representations.

The WO 2015/157410 A1 discloses a system and method for night vision object detection and driver assistance in which a stereo vision system contains two camera sensors, both of which are sensitive to infrared radiation and capture images of a scene. The camera sensors are connected to a processor that configures stereo tuning based on the signals emitted by both camera sensors.

From the US 2020/0177935 A1 a multispectral camera is known which includes several camera sensors in a housing, each camera sensor being sensitive to a different spectral range of light.

According to the US 2016/0071279 A1 It is known to compensate for inequalities in images taken by multiple cameras. A hybrid camera array has multiple camera sensors that have different resolutions and color characteristics. To compensate for the differences, the image from a camera sensor is compared with the image from a reference camera sensor.

The EP 3 723 364 A1 shows an image processing unit and a method for image processing using a hybrid camera comprising a visible light-sensitive camera and an infrared light-sensitive camera, which are alternately used to capture an image of a scene depending on whether it is day or night or is light or dark.

In the EP 3 659 002 B1 A vehicle interface for an autonomous vehicle is shown, in which the surroundings of the vehicle are recorded with cameras for further evaluation.

From the EP 3 477 540 A1 a method and a device for tracking an object are known, in which the object is tracked with a camera that is sensitive to visible or infrared light. These cameras can be combined into a hybrid camera. The camera can also be designed as a hybrid stereo camera.

According to the EP 2 353 298 B1 a method and a system for generating visual multi-view 3D content are known, in which bidimensional images of a scene are generated and evaluated from at least two views.

The hybrid stereo cameras described include two similar imaging units, i.e. both are RGB imagers or both grayscale imagers and both with or without filters for infrared light.

It is the object of the invention to overcome the disadvantages of the prior art and to provide a stereo vision system that can be used in as many applications as possible to reproduce a 3D scene.

This task is solved by the stereo vision system according to claim 1. Thereafter, a stereo vision system, which is designed to reconstruct a 3D representation of the scene from 2D representations of two monocular imaging units observing a scene by deriving depth data, comprises at least one monocular imaging unit that is sensitive to radiation outside a visible spectrum. This makes it possible to combine different types of imaging units with different sensitivity in terms of their spectral sensitivity. These imaging units include thermal imaging cameras as well as radar or lidar systems. This combination continues to enable stereo processing. At the same time, the operating conditions of the system are improved by making the cameras more sensitive to a wide range of lighting conditions. This increases the number of uses of the stereo vision system to display a 3D scene. Avoiding the use of two similar imaging units, such as two RGB or two infrared imaging units, results in less redundancy in the required stereo configuration.

The task is also solved by a driver assistance system for a vehicle, comprising a stereo vision camera system, which is designed to reconstruct a 3D representation of the scene from 2D representations of two monocular imaging units observing a scene by deriving depth data, with at least one monocular imaging unit is sensitive to radiation outside a visible spectrum. Driver assistance systems use the stereo vision camera system to record the surroundings of the vehicle in order to support the driver of the vehicle in his vehicle control tasks or to at least partially take over these. The more detailed the vehicle surroundings are using the stereo vision system can be displayed, the more precisely environmental data is determined and the more reliably and error-free the driver assistance system works in as many traffic situations as possible. The driver assistance system controls components of the vehicle in order to provide the assistance function, such as traffic sign recognition or automated, environment-based chassis adjustment. This means that dangerous situations in road traffic can be largely prevented. Based on the environmental data, an assistance function control signal is determined, with a control device accessing the respective predetermined functional context. This functional relationship includes at least one calculation rule that establishes a deterministic relationship between the environmental data and the assistance control signal and the assistance function to be achieved.

The environment captured by the stereo vision system is referred to below as a scene. This scene is imaged from different perspectives by at least two monocular imaging units. Each monocular imaging unit represents a separate unit and includes optics for capturing the scene. The section of the scene determined by the optics is directed to a sensor unit positioned in the imaging unit, which is designed as an imager for creating a 2D image of the captured scene.

In a preferred embodiment, each of the two monocular imaging units for recording 2D images of the same scene is connected to an image processing unit for outputting a 3D representation of the scene derived from the superimposed 2D images of the two monocular imaging units, and the image processing unit is for deriving the depth data of the observed scene from the superimposed 2D images. To superimpose the images from the two monocular imaging units, the electrical output signals of both imaging units, which are expressed as image sequences, are combined into a sequence. This sequence is fed to the image processing unit. This involves calibration in order to correct radiometric inequalities due to different light sensitivities of the two imaging units. This means that the data format (e.g. to 8-bit gray values), the brightness and the contrast are adapted. A 3D representation in the form of a 3D point cloud is then obtained from the calibrated sequence of successive images.

In a preferred embodiment, the first monocular imaging unit, which is sensitive to radiation outside a visible spectrum, is designed as an infrared camera, which is connected to a reconstruction unit of the image processing unit, which is designed to derive information for the 3D representation from the images generated with a time delay when the infrared camera moves . The image processing unit is formed by a processor which uses photogrammetric distance imaging methods to estimate three-dimensional structures from two-dimensional image sequences. One of these methods, carried out in the reconstruction unit, is known as structure from motion (SFM), which couples the three-dimensional structures with local motion signals. Due to a lateral movement of the images provided by the infrared camera around the scene to be recorded, objects in the scene move around it depending on their distance, which is referred to as motion parallax. Information from the depth of this motion parallax can be used to generate an accurate 3D representation of the scene. In addition to collecting depth information from the scene, 3D information can also be determined for a reliable representation of the 3D point cloud.

In a preferred embodiment, the image processing unit comprises an image fusion unit for stereo analysis, which is designed to adapt the 2D images of both monocular imaging units and to use them to determine depth data for different spatial points of the scene in order to reconstruct the 3D representation of the scene. In such a stereo analysis, the two 2D images involved are brought into an epipolar geometry. This means that the 2D images are rectified and adjusted so that a visible shift in the image due to different viewing angles is only present along the horizontal lines in the image pairs. The horizontal shift is inversely proportional to the depth of the assigned scene point. Standard algorithms such as semi-global matching or dynamic programming can be used to capture stereo correspondences.

In a preferred embodiment, the image fusion unit is coupled to the reconstruction unit to generate a 3D representation of the scene. This ensures that only the 2D images corrected for visible displacement are fed to the reconstruction unit of the 3D point cloud.

In a preferred embodiment, the two monocular imaging units have different imaging properties. Despite different imaging properties in different areas of the light spectrum The stereo capability of the imaging units working around is retained.

In a preferred embodiment, the monocular imaging units differ in their focal length, with the stereo analysis of the 2D images of the two monocular imaging units taking place in the field of view of the monocular imaging unit having the larger focal length. This has the advantage that the field of view of the imaging unit with the smaller focal length is already contained in the imaging unit with the larger focal length. The field of view is understood to be the field of view of the imaging units. In order to adjust the dimensions of the visibility, resampling of the images can be carried out.

In a preferred embodiment, the monocular imaging units differ in their resolutions. These different resolutions are compensated for in the stereo processing of the image fusion unit. This can be done by applying a scaling factor to at least one image before starting the search for stereo correspondences in the 2D images, i.e. before searching for correspondences between pixels of the two 2D images.

In a preferred embodiment, the second monocular imaging unit is designed as a camera that is sensitive to radiation within the visible spectrum, preferably as an RGB camera. The combination of imaging units that are sensitive to different spectral ranges allows the stereo vision system to be used day and night while providing color information. The stereo vision system can provide a full color image during the day and a grayscale image using the infrared camera in the dark. If the scene to be observed is sufficiently illuminated, depth information can be determined using the RGB camera even at night.

In a preferred embodiment, the second monocular imaging unit is designed as an imaging unit that is sensitive to radiation outside the visible spectrum, preferably as an infrared camera. The use of two infrared cameras that are sensitive to radiation in the non-visible spectrum allows a gray zone representation of the scene over a wide spectral range that includes the visible range and can therefore be used day and night.

In addition to focal length or resolution, other imaging properties can also be varied, such as the color sensitivity of the imaging units. To enable a comparison between two 2D images captured with different color filters, both 2D images can be brought into the same color space.

The aspects described here in relation to the stereo vision system apply equally to the disclosed driver assistance system, which consists of hardware and software within the vehicle. The hardware includes in particular digital signal processors, application-specific integrated circuits, field programmable gate arrays and other suitable switching and computing components.

The task is also solved by a method for generating a stereo vision from two 2D representations, in which a 3D representation of the scene is reconstructed from 2D representations of two monocular imaging units observing a scene by deriving depth data, with at least one monocular imaging unit for radiation is sensitive outside a visible spectrum. This combination enables stereo processing, i.e. collecting depth information about the observed scene. At the same time, operating conditions are improved by making the imaging units more receptive to a wide range of lighting conditions. This increases the number of applications of the stereo vision system to reconstruct a 3D scene.

In a preferred embodiment, the monocular imaging unit, which is sensitive to radiation outside a visible spectrum, is moved parallel to the scene and the 3D representation of the observed scene is derived from several 2D images generated with a time delay. For this purpose, photogrammetric distance imaging methods are used to estimate three-dimensional structures from two-dimensional image sequences. In addition to collecting depth information from the scene, 3D information can also be determined for a reliable representation of a 3D point cloud.

In a preferred embodiment, the 2D images of both monocular imaging units are adjusted and depth data for the various spatial points of the scene are determined from them to reconstruct the 3D representation of the scene. In a stereo analysis, the two 2D images involved are brought into an epipolar geometry, which represents a geometric calibration. This means that the 2D images are rectified and adjusted so that a visible shift in the image due to different viewing angles is only present along the horizontal lines in the 2D image pairs. The visible displacement is also called called disparity. This disparity is indirectly proportional to the depth of the associated point of the 3D representation.

• 1: eine schematische Darstellung eines ersten Ausführungsbeispiels des erfindungsgemäßen Stereovisions-Systems,
• 2 eine schematische Darstellung eines weiteren Ausführungsbeispiels des erfindungsgemäßen Stereovisions-Systems,
• 3 eine schematische Darstellung eines weiteren Ausführungsbeispiels des erfindungsgemäßen Stereovisions-Systems,
• 4 eine schematische Darstellung eines weiteren Ausführungsbeispiels des erfindungsgemäßen Stereovisions-Systems,
• 5 eine schematische Darstellung eines weiteren Ausführungsbeispiels des erfindungsgemäßen Stereovisions-Systems,
• 6 eine schematische Darstellung eines Ausführungsbeispiels zur monokularen Bildverarbeitung mit einer Infrarot-Kamera,
• 7 eine schematische Darstellung des erfindungsgemäßen Fahrerassistenzsystems.

Further features, advantages and properties of the invention are explained based on the description of preferred embodiments of the invention with reference to the figures, which show:

• 1 : a schematic representation of a first exemplary embodiment of the stereo vision system according to the invention,
• 2 a schematic representation of a further exemplary embodiment of the stereo vision system according to the invention,
• 3 a schematic representation of a further exemplary embodiment of the stereo vision system according to the invention,
• 4 a schematic representation of a further exemplary embodiment of the stereo vision system according to the invention,
• 5 a schematic representation of a further exemplary embodiment of the stereo vision system according to the invention,
• 6 a schematic representation of an exemplary embodiment for monocular image processing with an infrared camera,
• 7 a schematic representation of the driver assistance system according to the invention.

A first exemplary embodiment of a stereo vision system 1 according to the invention is shown in 1 shown schematically. The stereo vision system 1 consists of an RGB camera 2 and an infrared camera 3, the fields of view 4, 5 of which are directed at a scene 6 in an environment of the stereo vision system 1 and independently generate 2D images of the scene 6. RGB stands for the colors red, green and blue. The RGB camera 2 generates a 2D image 7 in color and the infrared camera 3 generates a 2D image 8 in grayscale. The respective images are converted into a separate electrical output signal in the corresponding camera 2, 3. The output signal 9 of the RGB camera 2 and the output signal 10 of the infrared camera 3 are combined into a binocular image sequence 11. The binocular image sequence 11 is fed as an input signal to an image processing unit 12, which is also part of the stereo vision system 1. The image processing unit 12, designed as a computing unit, includes an image fusion unit 13 for determining horizontal disparities D1, D2 between the two images 7, 8 of the RGB camera 2 and the infrared camera 3 and a reconstruction unit 14, which operates according to an algorithm for executing the “Structure from Motion " is trained. The output signal 15 of the image processing unit 12 enables a 3D point representation 16 of the scene 6 recorded by both cameras 2, 3.

The RGB camera 2 and the infrared camera 3 are positioned at a distance in a direction that is orthogonal to the target line vectors 17 of the two cameras 2, 3, which is also referred to as the baseline BL. The prerequisite here is that the target line vectors 17 of the two cameras 2, 3 are aligned parallel to one another. The baseline BL and the collinearity tolerance of the target line vectors 17 of the two cameras 2, 3 affect the three-dimensional accuracy.

In the image fusion unit 12, as in 1 shown as an example, the two camera images 7, 8 are brought into a so-called epipolar geometry. This means that the images 7, 8 are corrected and adjusted so that a parallax effect, ie the visible shift in the image due to different viewing angles, is only present along the horizontal lines in the image pairs. The depth of the associated points PSn of the recorded scene 6 is determined using two corresponding points in the images 7, 8 of the RGB camera 2 and the infrared camera 3. A point P71 of the image 7 of the RGB camera 2 is compared with the point P81 of the image 8 of the infrared camera 3 corresponding to the scene 6. A disparity D1 (visible shift) of the point P71 of the image 7 is determined with the target line vector 17 of the RGB camera 2. At the same time, a second disparity D2 is determined between the point P8n of the image 8 of the infrared camera 3 and the corresponding target line vector 17 of the infrared camera 3. If the same focal lengths BW of the two cameras 2, 3 are known, the depth TPS1 of the point PS1 of scene 6 is determined: $$\mathrm{<figure-callout\; id="1"\; label="T\; P\; S"\; filenames="DE102022123295A1\_0007.png,DE102022123295A1\_0009.png"\; state="\{\{state\}\}">T</figure-callout>}PS1=bW\ast bL/\left(\mathrm{<figure-callout\; id="1"\; label="D"\; filenames="DE102022123295A1\_0007.png,DE102022123295A1\_0009.png"\; state="\{\{state\}\}">D</figure-callout>}1+D2\right)$$

If the cameras have 2, 3 different focal lengths BW, the overlapping image sections must be focused in an intermediate processing step so that they correspond to the same focal lengths BW.

Analogous to this description, further points P7n of the image 7 of the RGB camera 2 are compared with the points P8n of the image 8 of the infrared camera 3 in order to determine the depth TPSn of the respective point PSn of the scene 6. $$TPSn=bW\ast bL/\left(D1n+D2n\right)$$

The 3D point representation 16 of the scene 6 is output by the image processing unit 12 from the large number of depth information of the individual scene points PS1 ... PSn determined in this way.

The reconstruction unit 14 is also contained in the image processing unit 12, which is designed as a computing unit. The “Structure from Motion” algorithm for determining further depth information is stored in the reconstruction unit 14. This algorithm is a photogrammetric evaluation method for estimating three-dimensional structures from two-dimensional image sequences that are coupled with local movement signals. In order to record the same image, the infrared camera 2 is moved parallel to the scene 6 to be captured, which creates time-shifted images from which 3D information is obtained. To find a correspondence between the two-dimensional images, selected points are traced from one image to the next. These point trajectories over time are used to reconstruct the 3D positions of the points and the movement of the camera, thereby determining the 3D point representation 16 of the scene 6 as a point cloud.

A schematic representation of a further exemplary embodiment of the stereo vision system 18 according to the invention is shown in 2 shown. This stereo vision system 18 differs from that in 1 stereo vision system 1 shown in that the RGB camera 2 is replaced by another infrared camera 19.

In this case too, the output signal 20 of the infrared camera 19 and the output signal 10 of the infrared camera 3 are combined and form a binocular infrared image sequence 21, which is fed to the image processing unit 12 as an input signal. The image processing unit 12 in turn includes the image fusion unit 13 and the reconstruction unit 14 by means of which depth information for the 3D point representation 16 is obtained in the manner described above from the point comparisons of the image 22 of the second infrared camera 19 and the image 8 of the first infrared camera 3 determined and issued.

3 shows a schematic representation of a further exemplary embodiment of the stereo vision system 23 according to the invention. This stereo vision system 23 corresponds in its basic structure to the stereo vision system 1 in 1 . The difference is that the output signal 9 of the RGB camera 2 and the output signal 10 of the infrared camera 3 are combined to form a stereo infrared RGB image sequence 24. This stereo infrared RGB image sequence 24 is fed as an input signal to the image fusion unit 13 of the image processing unit 12. In addition to converting the epipolar geometry, the image fusion unit 13 adapts the data format, for example with regard to the image resolution, brightness or contrast of the stereo infrared RGB image sequence 24. The stereo image pair is forwarded to the reconstruction unit 14, which outputs the 3D point cloud from this stereo image pair based on the “Structure from Motion” algorithm as a 3D representation 16 of the scene 6 for further processing.

One too 3 comparable processing takes place if, as in 2 shown, two infrared cameras 2, 19 are used in the supervision system 25. According to 4 the input signal for the image processing unit 12, which is combined from the output signals 8, 20 of the two infrared cameras 3, 19, represents a stereo infrared image sequence 26. This is first fed to the image fusion unit 13, where an unadulterated stereo signal is determined. This stereo signal is also fed to the reconstruction unit 14, which generates the 3D point representation 16 of the scene 6 using the “Structure from Motion” algorithm.

In 5 a schematic representation of a further exemplary embodiment of the stereo vision system 27 according to the invention is shown. In the stereo vision system 27, an unequal number of cameras are used for each side of the stereo configuration. An infrared camera 29, 30 is arranged on both sides of an RGB camera 28. This arrangement is carried out, for example, in such a way that the RGB camera 28 has a viewing area 32 of 180°, while the two infrared cameras 29, 30 each cover a viewing area 31, 33 of 90°. Both viewing areas 31, 33 of the infrared cameras 29, 30 cover the viewing area 32 of the RGB camera 28. However, the arrangement is not limited to these viewing area angles. A variety of other viewing angles can be selected.

From this arrangement, two pairs of stereo cameras 28, 29, 28, 30 can be constructed, i.e. the RGB camera 28 can each form a stereo camera pair with an infrared camera 29, 30. The output signals of the RGB camera 28 and an infrared camera 29 or 30 are each combined to form a stereo infrared and RGB image sequence 34, 35. Thus, two stereo, infrared and RGB image sequences 34, 35 form the input signals of the image processing unit 12. In contrast, the RGB camera 28 and the two infrared cameras 29, 30 can be used together to provide input for the “Structure from Motion”. In this case, a sequence of output signals from the RGB camera 28 and the two infrared cameras 29, 30 is formed as the input signal of the image processing unit 12 in order to make an estimate of the 3D point representation 16.

In addition to stereo processing, a single infrared camera 36 can also be used for monocular image processing to derive depth information using the structure from motion algorithm. This will be in 6 shown. The infrared camera 36 outputs a monocular infrared image sequence 37, which is forwarded to the image processing unit 12 as an input signal. In the image processing unit 12, only the reconstruction unit 14 is required, which includes the “Structure from Motion” algorithm in order to generate a 3D point representation 16 of the scene 6 in the manner described.

The use cases related to 2 and 3 have been explained, can be simplified by using the camera pairs 3, 19 or 2, 3 to carry out stereo triangulation, whereby the “Structure from Motion” algorithm can be dispensed with. With such a triangulation method, distance is determined by precise angle measurement of triangles. Trigonometric functions are used for the calculation.

A driver assistance system 38 according to the invention, as shown in 7 is shown, includes a control device 39, which is connected to the stereo vision system 1, 18, 23, 25, 27 used as environmental sensors. The driver assistance system 38 is installed in a vehicle 40, on which the stereo vision system 1 is also positioned. The driver assistance system 38 carries out one or more assistance functions that partially or completely relieve a driver of the vehicle 40 from the task of driving the vehicle 40.

For the sake of clarity, reference will only be made to the stereo vision system 1. The further description also applies to the other stereo vision systems. The stereo vision system 1 records environmental data of the vehicle 40. This environmental data can include images of objects, object positions, object edges, object distances, etc. be. Both static traffic-relevant components of road traffic, such as road markings, traffic signs or road surfaces, as well as moving ones, such as people and others, are considered objects.

In the present case, the driver assistance system 38 is designed, for example, as a traffic sign recognition system, with the stereo vision system 1 recording a traffic sign 41 in its surroundings. Due to temporary obscuration of traffic signs, e.g. E.g. due to other road users or plant growth, traffic signs can be poorly or not recognized by a camera. The described image processing methods of the stereo vision system 1 are used to localize the detected traffic sign 41 closer to its surroundings. The control device 39 includes a comparison unit 42 for comparing the extracted traffic sign 41 with reference traffic signs stored in a memory 43. Depending on this comparison, an assistance control signal for outputting or not outputting the traffic sign 41 on a display unit 45 of the vehicle 40 is generated for a display logic 44.

Reference symbols

1

Stereo vision system

2

RGB camera

3

Infrared camera

4

Camera field of view

5

Camera field of view

6

scene to be recorded

7

Picture

8th

Picture

9

RGB camera output signal

10

Infrared camera output signal

11

binocular image sequence

12

Image processing unit

13

Image fusion unit

14

Reconstruction unit

15

Output signal of the image processing unit

16

3D point cloud

17

Finish line vector

18

Stereo vision system

19

Infrared camera

20

Infrared camera output signal

21

binocular infrared image sequence

22

Image from the infrared camera

23

Stereo vision system

24

Stereo infrared RGB image sequence

25

Stereo vision system

26

Stereo infrared image sequence

27

Stereo vision system

28

RGB camera

29

Infrared camera

30

Infrared camera

31

Field of view infrared camera

32

Field of view RGB camera

33

Field of view infrared camera

34

Stereo and RGB image sequence

35

Stereo and RGB image sequence

36

Infrared camera

37

monocular infrared image sequence

38

Driver assistance system

39

Control unit

40

vehicle

41

Traffic sign

42

Comparison unit

43

Storage

44

Display logic

45

Display unit

BL

Baseline

BW

focal length

D1

Disparity

D2

Disparity

TPS

Depth data

P.S

Scene point

QUOTES INCLUDED IN THE DESCRIPTION

This list of documents listed by the applicant was generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Cited patent literature

• WO 2015157410 A1 [0002]
• US 20200177935 A1 [0003]
• US 20160071279 A1 [0004]
• EP 3723364 A1 [0005]
• EP 3659002 B1 [0006]
• EP 3477540 A1 [0007]
• EP 2353298 B1 [0008]

Claims (24)

Stereo vision system, which is designed to create a 3D representation (16) of the scene (6) from 2D representations (7, 8) of two monocular imaging units (2, 3) observing a scene (6) by deriving depth data (TPS). to reconstruct, wherein at least one monocular imaging unit (3) is sensitive to radiation outside a visible spectrum. Stereo vision system Claim 1 , each of the two monocular imaging units (2, 3) for recording 2D images (7, 8) of the same scene (6) with an image processing unit (12) for outputting one of the superimposed 2D images (7, 8). two monocular imaging units (2, 3) derived 3D representation (16) of the scene (6) are connected and the image processing unit (12) for deriving the depth data (TPS) of the observed scene (6) from the superimposed 2D images (7, 8) is trained. Stereo vision system Claim 1 or 2 , wherein the first monocular imaging unit, which is sensitive to radiation outside of a visible spectrum, is designed as an infrared camera (3, 37), which is connected to a reconstruction unit (14) of the image processing unit (12), which is designed from which, when the infrared camera ( 3, 37) to derive information for the 3D representation (16) from images generated with a time delay. Stereo vision system according to at least one of the preceding claims, wherein the image processing unit (12) comprises an image fusion unit (13) for stereo analysis, which is designed to adapt the 2D images (7, 8) of both monocular imaging units (2, 3) and from them To determine depth data (TPS) for different spatial points (PS) of the scene (6) to reconstruct the 3D representation (16) of the scene (6). Stereo vision system according to at least one of the preceding claims, wherein the image fusion unit (13) is coupled to the evaluation unit (14) to generate a 3D representation (16) of the scene (6). Stereo vision system according to at least one of the preceding claims, wherein the image fusion unit (13) is designed to correct radiometric inequalities in the monocular imaging units (2, 3). Stereo vision system according to at least one of the preceding claims, characterized in that the two monocular imaging units (2, 3) have different imaging properties. Stereo vision system according to at least one of the preceding claims, wherein the monocular imaging units (2, 3) differ in their focal length (BW), and the stereo analysis of the 2D images (7, 8) of the two monocular imaging units (2, 3) in Field of view of the monocular imaging unit (2, 3) having the larger focal length (BW). Stereo vision system according to at least one of the preceding claims, wherein the monocular imaging units (2, 3) differ in their resolutions. Stereo vision system according to at least one of the preceding claims, wherein the second monocular imaging unit is designed as a camera (2) sensitive to radiation within the visible spectrum, preferably as an RGB camera. Stereo vision system according to at least one of the preceding claims, wherein the second monocular imaging unit is designed as a camera sensitive to radiation outside the visible spectrum, preferably as an infrared camera (20). Driver assistance system for a vehicle, comprising a stereo vision camera system (1), which is designed to create a 3D from 2D representations (7, 8) of two monocular imaging units (2, 3) observing a scene (6) by deriving depth data (TPS). -Representation (16) of the scene (6) to be reconstructed, with at least one monocular imaging unit (3) being sensitive to radiation outside a visible spectrum. Driver assistance system Claim 12 , each of the two monocular imaging units (2, 3) for recording 2D images (7, 8) of the same scene with an image processing unit (12) for outputting one of the superimposed 2D images (7, 8) of the two monocular imaging units (2, 3) derived 3D representation (16) of the scene (6) are connected and the image processing unit (12) is designed to derive the depth data (TPS) of the observed scene (6) from the superimposed 2D images (7, 8). is. Driver assistance system Claim 12 or 13 , wherein the first monocular imaging unit, which is sensitive to radiation outside a visible spectrum, is designed as an infrared camera (3), which is connected to a reconstruction unit (14) of the image processing unit (12), which is designed to use the information when the information is moved Rared camera (3) to derive information for the 3D representation (16) from images generated with a time delay. Driver assistance system according to at least one of the preceding claims, wherein the image processing unit (12) comprises an image fusion unit (13) for stereo analysis, which is designed to adapt the 2D images (8, 22) of both monocular imaging units (2, 3) and to produce depth data ( TPS) for different spatial points (PSn) of the scene (6) to reconstruct the 3D representation (16) of the scene (6). Driver assistance system according to at least one of the preceding claims, wherein the image fusion unit (12) is coupled to the reconstruction unit (14) to generate a 3D representation (16) of the scene (6). Driver assistance system according to at least one of the preceding claims, wherein the two monocular imaging units (2, 3) have different imaging properties. Driver assistance system according to at least one of the preceding claims, wherein the monocular imaging units (2, 3) differ in their focal length (BW), the stereo analysis of the 2D images (7, 8) of the two monocular imaging units in the field of view of the larger focal length (BW ) having monocular imaging unit (2, 3). Driver assistance system according to at least one of the preceding claims, wherein the monocular imaging units (2, 3) differ in their resolutions. Driver assistance system according to at least one of the preceding claims, wherein the second monocular imaging unit is designed as a camera sensitive to radiation within the visible spectrum, preferably as an RGB camera (2). Driver assistance system according to at least one of the preceding claims, wherein the second monocular imaging unit is designed as a camera sensitive to radiation outside the visible spectrum, preferably an infrared camera (3). Method for generating a stereo vision from two 2D representations, in which a 3D representation (16) of the scene is created from 2D representations (7, 8) of two monocular imaging units (2, 3) observing a scene by deriving depth data (TPS). 6) is reconstructed, with at least one monocular imaging unit (3) being sensitive to radiation outside a visible spectrum. Procedure according to Claim 21 , wherein the monocular imaging unit (3), which is sensitive to radiation outside a visible spectrum, is moved parallel to the scene (6) and the 3D representation (16) of the observed scene (6) is derived from several 2D images generated with a time delay. Procedure according to Claim 21 or 22 , wherein the 2D images (8, 22) of both monocular imaging units (3, 20) are adapted and depth data (TPS) for different spatial points (PSn) of the scene (6) are used to reconstruct the 3D representation (16) of the scene ( 6) can be determined.

Images