METHOD FOR DETERMINING A BUCKLE ANGLE, PROCESSING UNIT AND VEHICLE

Description

[0001] The invention relates to a method for determining a bend angle, a processing unit for carrying out the method and a vehicle with the processing unit.

[0002] Estimating the articulation angle between a towing vehicle and a trailer coupled to it is necessary, for example, for stability functions or driver assistance functions. The articulation angle can be estimated, for example, by determining the trailer's pose or orientation in three-dimensional space, which is conventionally performed using appropriate sensors on the towing vehicle and / or image processing. For example, it is planned to spatially localize a trailer using LIDAR sensors or 3D stereo cameras and / or to estimate the trailer position and / or orientation using additional flat maker structures (QR codes or Aruco markers) on the trailer.It is also known how, with the help of a monocular camera, the structure of the scene can be determined in 3D using photogrammetric methods based on the forward or backward movement of a vehicle on which the camera is mounted (so-called Structure from Motion (SfM)).

[0003] For example, DE 10 2016 011 324 A1, DE 10 2018 114 730 A1, WO 2018 / 210990 A1, or DE 10 2017 119 968 A1 describe how to derive a trailer position and a trailer orientation of the respective object or trailer relative to the camera or towing vehicle using image processing. Depending on this, a bending angle or a distance, for example, can be determined. The trailer pose can also be determined with a mono camera by locating at least three markers, which are preferably applied flatly to the trailer, in a recorded individual image. Knowing the marker positions on the trailer, a transformation matrix is ​​determined from which the trailer pose can be derived.

[0004] Detection of a kink angle via a camera is further described in US 2014 200759 A, whereby a flat marker on the trailer is observed from the towing vehicle over time, and the kink angle is estimated from this. JP 2002 012 172 A, US2014277942A1, US 2008 231 701 A, and US 2017106796 A also describe kink angle detection based on flat markers. In US 2006 293 800 A, coupling points can carry a marker to facilitate automatic detection. The marker can have a special color, texture, or wave reflection property. DE 10 2004 025 252 B4 further describes determining the kink angle by sending radiation from a transmitter onto a semicircular or hemispherical reflector and then detecting the radiation reflected by it. DE 103 025 45 A1 also describes how to detect a clutch using an object recognition algorithm.EP 3 180 769 B1 also describes using cameras to capture the rear of a trailer and identify trackable features, such as an edge or corner, from the image. These are then tracked over time to determine, in particular, the articulation angle between the towing vehicle and the trailer.

[0005] US 2018 039 266 A additionally describes the use of information from a two-dimensional barcode or QR code. DE 10 2016 209 418 A1 also describes the use of a QR code with a camera to identify the trailer and transmit trailer parameters to a reversing assistance system. Alternatively or additionally, an RFID reader on the trailer can read an RFID transponder attached to the towing vehicle, for example in the form of a label. This allows the position of the QR code or RFID transponder to be calculated. Orientation is not determined in this process. WO18060192A1 also proposes a solution using radio-based transponders. DE 10 2006 040 879 B4 also uses RFID elements on the trailer for triangulation when approaching.

[0006] WO 2008064892 A1 provides an observation element that has at least three auxiliary points or measurement points that can be detected by a camera. From geometric considerations, the coordinates or vectors of the centers of gravity of the auxiliary points are determined, and from these, the coordinates of the measurement points relative to the camera or image sensor are determined. From this, a bending angle can be determined.

[0007] The disadvantages of the described solutions are that these methods and systems are either very complex to implement, rely on costly sensors and additional markers or similar devices, or they cannot be reliably detected under different environmental conditions. For example, flat markers cannot be reliably detected in the dark and / or under extreme or shallow blinking angles, making it impossible to reliably determine the trailer's pose or articulation angle relative to the towing vehicle. Furthermore, trailers without such markers cannot be located.

[0008] DE 10 2008 045 436 A1 further proposes capturing the trailer in a single image and comparing this single image with a stored reference image. Each reference image shows the trailer at a different articulation angle relative to the towing vehicle, so that if the individual image and the reference image match, the articulation angle can be estimated. Therefore, to determine the articulation angle, a reference image must be captured for each articulation angle.

[0009] DE 10 2010 006 521 A1 further describes determining the articulation angle using flow vectors, with the flow vectors being determined from chronologically successive individual images. In DE 10 2016 121 514 A1, the articulation angle is calculated from the position of a rear edge of the trailer estimated by image processing. In DE 10 2019 103 742 A1, the articulation angle is determined from sensor data using two different algorithms. In DE 10 2019 106 275 A1, a pattern is projected onto a partial area of ​​a partial vehicle, and the articulation angle is determined from the change in the imaged pattern. In DE 10 2019 110 018 A1 and DE 10 2019 115 789 A1, the articulation angle is recorded using a sensor, such as an articulation angle sensor or a camera.Also in DE 10 2018 005 118 A1, EP 2 551 132 B1 or DE 10 2012 006 206 A1, the determination of the bending angle can be carried out via a camera, in particular a monocular camera, using a structure-from-motion (SfM) method.

[0010] US 2020 143174 A1 discloses a system and method for detecting trailer angles. In one embodiment, an in-vehicle control system includes an optical sensor configured to be mounted on a tractor so that it is directed toward a trailer coupled to the tractor, the optical sensor further configured to generate optical data indicative of an angle formed between the trailer and the tractor.The system further includes a processor and computer-readable memory in communication with the processor and having computer-executable instructions stored thereon to cause the processor to receive the optical data from the optical sensor, determine at least one candidate plane representing a surface of the trailer visible in the optical data based on the optical data, and determine an angle between the trailer and the tractor based on the at least one candidate plane.

[0011] The object of the invention is to provide a method for determining the articulation angle between a towing vehicle and a trailer coupled to it, with which the articulation angle can be determined simply and reliably. Furthermore, the object is to provide a processing unit and a vehicle for implementing the method.

[0012] This object is achieved by a method, a processing unit, and a vehicle according to the independent claims. The subclaims specify preferred developments.

[0013] According to the invention, a method is provided for determining a bending angle between a towing vehicle and a trailer coupled thereto, wherein the trailer can be imaged in individual images with individual image pixels using a camera system, in particular with a monocular camera, wherein at least the following steps are provided: Determining a point cloud from at least one individual image in which at least the trailer is depicted, wherein the point cloud consists of object points that are assigned to the trailer. The point cloud is preferably present in a three-dimensional coordinate system in which the individual object points are represented via their world coordinates. The point cloud can be created from the at least one recorded individual image using image processing methods, preferably using a structure-from-motion (SfM) method from at least two individual images captured by a monocular camera or from a single image from a stereo camera using appropriate image processing methods. The camera used can advantageously be an existing or otherwise usable reversing camera whose field of view is aligned or can be aligned to the rear of the towing vehicle, in which the trailer is normally located.This could, for example, be a camera designed to monitor the area behind the towing vehicle in compliance with UNE-CE regulations. Form at least one flat or level reference surface from those neighboring object points in the point cloud that lie essentially in the same plane, with the respective flat reference surface having a fixed spatial orientation and / or a fixed spatial position relative to the trailer. This involves searching for neighboring object points in the point cloud that lie in the same plane and form a contiguous surface on the trailer, and assigning these to the flat reference surface.In order to be able to access this subsequently, the at least one formed flat reference surface and / or the underlying point cloud can be saved, preferably together with the spatial orientation and / or the spatial position of the respective flat reference surface relative to the trailer, in order to be able to access information about the spatial position subsequently. Reading in at least one individual image in which the trailer coupled to the towing vehicle is shown, and determining at least one partial surface in the at least one read-in individual image. In the coupled state, a subsequent search is carried out again in the individual image for partial surfaces on the trailer. Fitting the at least one previously formed and stored flat reference surface into the at least one partial surface determined from the individual image in the coupled state using a geometric transformation. The template orA flat reference surface serving as a template, the relative position (reference orientation / reference position) of which on or relative to the trailer is known, is then fitted into the partial surface of the depicted trailer in the current driving situation. The articulation angle between the towing vehicle and the trailer is determined as a function of a rotational component of the applied geometric transformation, whereby the rotational component of the applied geometric transformation characterizes a rotation of the flat reference surface about a swivel axis of the trailer. From the transformation (rule) that maps the previously determined template or the reference surface to the partial surface from the individual image of the current driving situation, only the relevant rotational component is used to estimate the changed relative position of the reference surface and thus also of the trailer and from this the articulation angle.

[0014] Advantageously, a template matching method enables a simple estimation of the articulation angle, whereby only a limited surface area of ​​the trailer needs to be considered, which preferably has a simple geometric shape, for example, a quadrilateral. This surface area does not necessarily have to be applied to the trailer beforehand, as one or more already existing, definable surface areas can be used. Such surface areas are automatically searched for on the trailer in advance to determine the template or reference area, and are then retrieved accordingly during the journey, so that the articulation angle can be easily estimated from geometric considerations using the geometric transformation.In the simplest case, a camera system with a camera directed toward the rear and a corresponding image processing algorithm are sufficient to clearly determine the surface areas. By using an SfM algorithm (SfM: Structure-from-Motion), a monocular camera is advantageously sufficient, which minimizes costs and processing effort.

[0015] According to the invention, a processing unit for carrying out the method according to the invention and a vehicle with such a processing unit are also provided.

[0016] Preferably, it is further provided that the at least one individual image for determining the point cloud is recorded when the trailer is not yet coupled to the towing vehicle, for example, during a coupling process. The reference area is thus determined based on a point cloud recorded in a driving situation in which the towing vehicle is highly likely to be moving relative to the trailer. This allows corresponding reference areas to be extracted from the point cloud in a clear and reliable manner, since the trailer is reliably and ideally also comprehensively captured in its three-dimensional form in the point cloud, particularly when the point cloud is determined using an SfM algorithm.

[0017] Preferably, the point cloud is determined using the SfM algorithm from at least two individual images, with the trailer being captured in the at least two individual images from at least two different viewpoints, preferably by a monocular camera of the camera system. This allows a point cloud to be formed from multiple object points, which are at least also associated with the trailer, and depth information about the respective object points can be obtained from the at least two individual images through triangulation. Accordingly, only two individual images are required to generate the point cloud.

[0018] Preferably, it is further provided that the at least one flat reference surface is defined from adjacent object points that lie in the same plane in such a way that the flat reference surface can be assigned to a color- and / or at least structurally delimited or delimitable region of adjacent individual image pixels in the at least one individual image. Accordingly, the flat reference surface is defined not only based on its position in a plane, but also in such a way that it can be clearly identified in a two-dimensionally recorded image. This simplifies the fitting process and allows it to be carried out more reliably, since the reference surface can subsequently be clearly identified.

[0019] Preferably, it can be provided that the flat reference surface is formed from those neighboring object points of the point cloud that lie in the same plane and / or to which individual image pixels with the same or at least similar color values ​​and / or intensity values ​​are assigned in the at least one individual image. In this way, a partial surface that can be demarcated in terms of color and / or structure can be used. Accordingly, it can be provided that the at least one flat reference surface completely covers the front of a trailer if, for example, no smaller partial surfaces that can be demarcated in terms of color and / or structure can be found on the front of the trailer from which a clearly demarcated reference surface can be derived, or if it only partially occupies the front of the trailer. In this way, a different partial surface orA reference area can be searched for and defined, which means that the process can be used flexibly and does not depend on specific patterns or markers on the trailer.

[0020] Preferably, it is further provided that neighboring individual image pixels with the same or at least similar color values ​​and / or intensity values ​​are assigned to the same partial area of ​​the scanned individual image. In this way, partial areas can be easily located based on the color or intensity in the image of the tag.

[0021] Preferably, it is further provided that an intermediate surface is formed from the at least one previously formed flat reference surface by means of a coordinate transformation, wherein the flat reference surface is present in a three-dimensional coordinate system and the intermediate surface in a two-dimensional coordinate system, and the respectively formed intermediate surface is fitted into the respective partial surface of the read-in individual image by means of the geometric transformation. Therefore, two processing steps are provided if the reference surface and the partial surface are present in different dimensions. In order to be able to carry out the geometric transformation or fitting into the partial surface in this case, a coordinate transformation must also be applied, wherein the geometric and coordinate transformations can be carried out together by applying a common transformation matrix.

[0022] Preferably, the intermediate surface formed is shifted and / or rotated and / or sheared and / or scaled by the geometric transformation, for example, in multiple iterative steps. An iterative process allows the fitting process to be carried out simply and reliably, with the resulting geometric transformation then allowing the bending angle to be determined from geometric considerations. To account for different distances or sizes of the reference surface, the search or fitting can take place at different resolution levels (keyword: image pyramid).

[0023] Preferably, it is further provided that pixel-by-pixel color values ​​and / or intensity values ​​in the respective partial area and in the formed intermediate area are compared to fit the at least one previously formed flat reference area into the at least one determined partial area. The fitting can thus be carried out in a simple and reliable manner by searching for pixel similarities between the respective areas, for example, in the iterative process.

[0024] The invention is explained in more detail below with reference to the figures. They show: Fig. 1 shows schematic views of a vehicle; Fig. 2 shows a flowchart of the method according to the invention; Fig. 3 shows a detailed view of an object imaged in two individual images from a single camera; and Fig. 4 shows a detailed view of the individual images recorded.

[0025] Figure 1shows a schematic plan view of a vehicle 1 consisting of a towing vehicle 1a and a trailer 1b, wherein the trailer 1b is coupled to the towing vehicle 1a in the situation shown. Using a method according to the invention, which is described below, a bending angle KW between the trailer 1b and the towing vehicle 1a, ie, an angle between the longitudinal axis of the trailer and the longitudinal axis of the towing vehicle, can be determined.

[0026] For this purpose, a camera system 2 is arranged on the towing vehicle 1a, which has at least one camera 2a, in particular in a monocular design, with which an environment U, in particular behind the towing vehicle 1a, can be recorded. A field of view 4 of the at least one camera 2a is therefore directed in particular towards a rear space R behind the towing vehicle 1a. In order to be able to cover the entire or at least a large area of ​​the environment U, in particular the rear space R, the camera(s) 2a can be designed, for example, as fisheye cameras, each of which can cover a field of view 4 with a viewing angle of equal to or greater than 170°.

[0027] Image signals SB output by the respective camera 2a are output, if necessary pre-processed, to a processing unit 5, e.g. in the towing vehicle 1a, which is designed to process individual images EBk of the respective camera 2a based on the image signals SB in order to Fig. 2to carry out the method shown as an example for determining the articulation angle KW between the towing vehicle 1a and the trailer 1b: The method is divided into two parts, with preliminary method steps VST1 to VST6 being carried out, for example during a coupling process AV, and main method steps ST1 to ST5 being carried out while the vehicle 1 is operated as a combination consisting of the towing vehicle 1a and the coupled trailer 1b, i.e. a specific articulation angle KW is established between the two. The preliminary method steps VST1 to VST6 described below exploit the fact that the towing vehicle 1a with the camera 2a is always moving relative to the trailer 1b. As a result, by using an SfM algorithm AS (SfM: Structure from Motion) the processing unit 5 can generate a 3D reconstruction of the current scene from at least two individual images EBk recorded by the camera 2a.

[0028] In a first preliminary process step VST1, two (k = 1, 2) or more (k = 1, 2, 3, ...) individual images EBk characterizing the environment U around the towing vehicle 1a are first read in by the processing unit 5, with the trailer 1b to be coupled being depicted as object O in the individual images EBk. The individual images EBk contain a 2D representation of the current scene (scene representation in two dimensions).

[0029] Subsequently, in a second pre-process step VST2, depth information Tln is extracted from these at least two individual images EBk using the SfM algorithm AS for object points POn recorded or depicted in the individual images EBk (see Fig. 3), which are assigned at least to the trailer 1b in the rear space R. The extraction of depth information Tln via the SfM algorithm AS is carried out by recording an object point POn on the trailer 1b from at least two different positions SP1, SP2 by the same camera 2a, as in Fig. 3 indicated.

[0030] Through triangulation T, the depth information Tln can then be obtained for each pair of object points POn. For this purpose, image coordinates xB, yB are calculated in a two-dimensional coordinate system K2D (see Fig. 4) to at least one first individual image pixel EB1P1 in a first individual image EB1 and at least one first individual image pixel EB2P1 in a second individual image EB2 of the same camera 2a. Both individual images EB1, EB2 are recorded by the camera 2a at different viewpoints SP1, SP2, ie between the individual images EB1, EB2 the towing vehicle 1a or the camera 2a moves by a base length L. The two first individual image pixels EB1P1, EB2P1 are selected in the respective individual images EB1, EB2 in a known manner such that they are assigned to the same object point POn on the respective trailer 1b depicted.

[0031] In this way, for a trailer 1b with its object points POn, one or more pairs of individual image pixels EB1Pi, EB2Pi can be determined for one (n=1) or more (n=1, 2, ...) object points POn. As an approximation, the absolute, actual object coordinates xO, yO, zO (world coordinates) of the respective object point(s) POn of the three-dimensional trailer 1b can then be calculated or estimated from the image coordinates xB, yB of the individual image pixels EB1Pi, EB2Pi determined for the respective trailer 1b or the object points POn by triangulation T. In order to be able to carry out the triangulation T, a correspondingly determined base length L between the viewpoints SP1, SP2 of the camera 2a is used, for example from movement data of the towing vehicle 1a or the camera 2a.

[0032] If motion data is not available, it is also possible to determine the motion data using visual odometry, i.e., "visually" from the individual images EBk, within the SfM algorithm AS. For this purpose, the movement of camera 2a is estimated based on temporally tracked feature points or object points POn from individual images EBk.

[0033] If the triangulation T has been performed for a sufficient number of object points POn of a trailer 1b, a three-dimensional point cloud PW consisting of several object points POn in actual object coordinates xO, yO, zO (world coordinates) can be generated in a third preliminary process step VST3 in order to describe the respective trailer 1b in three-dimensional space. Using the SfM algorithm AS, a point cloud PW is generated in three-dimensional space from the two-dimensional representation of the scene (scene representation in the individual images EBk), which, for example, contains side surfaces of the trailer 1b. To verify that these are actually object points POn on the trailer 1b, the point cloud PW can, for example, be compared with a stored trailer model from a trailer database.

[0034] In a fourth preliminary process step VST4, one or more definable flat sub-areas on the trailer 1b are searched for in the point cloud PW. This is done, for example, by geometrically examining neighboring object points POn or contiguous subsets of the point cloud PW to determine whether they lie in a plane E. If several object points POn of the point cloud PW that are adjacent to one another are arranged within the same plane E, taking a tolerance into account, a contiguous flat reference surface RF is formed from these neighboring object points POn.

[0035] Preferably, the flat reference surface RF is additionally defined in such a way that it is also delimited in the individual images EBk, for example, by color and / or structure, so that the flat reference surface RF can also be found in the two-dimensional individual images EBk. For the respective object points POn that lie in the same plane E, a supplementary check may be performed to determine whether the same or similar color values ​​CW and / or intensity values ​​IW are assigned to them in the individual images EBk.

[0036] The flat reference surface RF can, for example, extend over the entire trailer front 1c, e.g., if there are no color and / or structural differences on the trailer front 1c from which a clearly definable reference surface RF can be derived, or over partial areas of the trailer front 1c that are correspondingly delimited by color and / or structure, for example, by shading. Multiple flat reference surfaces RF can also be found if connected object points POn can be found or identified in the same plane E in the point cloud PW.

[0037] In a fifth preliminary process step VST5, a spatial reference orientation RFO and / or a spatial reference position RFP (or the reference pose) relative to the trailer 1b is determined for each determined planar reference surface RF in order to determine its exact position in space (in 3D). The reference orientation RFO and the reference position RFP can, for example, be specified relative to the kingpin of the trailer 1b or a corner of the trailer 1b, whereby this can be derived directly from the object coordinates xO, yO, zO of the object points POn assigned to the respective planar reference surface RF and / or from the known geometric dimensions of the respective trailer 1b.

[0038] Each found flat reference surface RF is then stored in a sixth preliminary process step VST6 with its corresponding spatial reference orientation RFO relative to the trailer 1b and, if applicable, its spatial reference position RFP relative to the trailer 1b (or the pose), e.g., as part of a reference representation FDR. This reference surface(s) RF or this reference representation FDR serves subsequently as a pattern or template for determining a bending angle KW by determining how the position of the reference surface(s) RF changes during the travel of the vehicle 1 relative to the towing vehicle 1a.

[0039] For this purpose, in a first main method step ST1, the stored reference surface(s) RF or the reference representation FDR is / are read in by the processing unit 5, while the trailer 1b, for which the preliminary method steps VST1 to VST6 were previously carried out, is coupled to the towing vehicle 1a, which can be checked accordingly beforehand in an initialization step ST0.

[0040] Subsequently, in a second main method step ST2, at least one individual image EBk from camera 2a is read by the processing unit 5 while the towing vehicle 1a is operating with the coupled trailer 1b. In a third main method step ST3, one or more partial areas FT are identified in the at least one read individual image EBk, whereby, for example, neighboring individual image pixels EBkPi with the same or similar color values ​​CW and / or intensity values ​​IW are assigned to the same partial area FT.

[0041] If one or more partial surface(s) FT have been identified in this way in the at least one individual image EBk, an attempt is made in a fourth main method step ST4 to fit the read-in planar reference surface(s) RF into this partial surface(s) FT, for example in several iteration steps STI. In this case, the read-in planar reference surface(s) RF, which are present spatially or in a three-dimensional coordinate system KD3, is / are first converted into a two-dimensional coordinate system K2D by means of a coordinate transformation TK. As a result, two-dimensional intermediate surfaces RFZ are formed or generated from the planar reference surface(s) RF in order to represent the planar reference surface(s) RF (as intermediate surfaces RFZ) in the same dimension as the partial surface(s) FT.By means of a geometric transformation TG, the generated intermediate surface(s) RFZ are then shifted and / or rotated and / or sheared and / or scaled, for example in an iterative process, in such a way that they fit into the partial surface(s) FT as best as possible.

[0042] For example, pixel similarities between a partial surface FT and a generated intermediate surface RFZ are sought to achieve a correct assignment. This means that, in the partial surface FT and the generated intermediate surface RFZ, the respective color values ​​CW and / or intensity values ​​IW are compared pixel by pixel. If a sufficient match is found, a correct assignment between a partial surface FT and an intermediate surface RFZ (or flat reference surface RF) is assumed.

[0043] From the geometric transformation TG, via which the respective intermediate surface RFZ is fitted into the respective partial surface RF, the articulation angle KW can then be directly derived from geometric considerations in a fifth main process step ST5. The geometric transformation TG transfers the flat reference surface RF transformed into the two-dimensional coordinate system K2D or the intermediate surface RFZ generated thereby, to which the reference orientation RFO determined in the fifth preliminary process step VST5 is assigned, to the partial surface FT on the trailer 1b. A rotational component TGR of the geometric transformation TG therefore describes in particular a rotation of the generated intermediate surface RFZ around a pivot axis H of the trailer 1b (cf. coupling point between trailer 1b and towing vehicle 1a), from which the change in the position of the trailer 1b relative to the towing vehicle 1a orto camera 2a compared to the state in the fifth pre-process step VST5.

[0044] It is assumed that the partial surface FT has the same reference orientation RFO relative to the trailer 1b as the respectively assigned reference surface RF, since its relative position to the trailer 1b does not change with varying articulation angle KW, but only the relative position to the towing vehicle 1a or the camera 2a. From geometric considerations, the change in the position of the trailer 1b relative to the towing vehicle 1a or the camera 2a then directly results in a change in the articulation angle dKW and, from this, via a corresponding pre-performed calibration, also in the articulation angle KW itself.

[0045] For each captured image EBk, the current bending angle KW can be determined using the flat reference surface(s) RF. If this is performed for multiple reference surfaces RF in parallel, a plausibility check can also be performed by comparing the bending angles KW derived from different reference surfaces RF. List of reference symbols (part of the description)

[0046] 1Vehicle 1aTowing vehicle 1bTrailer 1cTrailer front 2Camera system 2aCamera 4Field of view 5Processing unit ASSfM algorithm AVCoupling process CWColor values ​​dKWChange in articulation angle EPlane EBkk. Single image EBkPii. Single image - pixel in the kth single image EBk FDRReference representation FTPartial surface HSwidth axis IWIntensity value K2DTwo-dimensional coordinate system K3DThree-dimensional coordinate system KWArticulation angle LBase length OObject POnn. Object point PWPoint cloud RRear space RFPlane reference surface RFOReference orientation of the planar reference surface RF RFPReference position of the planar reference surface RF RFZZIntermediate surface SBImage signal SP11. Viewpoint SP22. Viewpoint STIIteration steps TTriangulation TGgeometric transformation TGRrotational part of the geometric transformation TG TlnDepth information TKCoordinate transformation UEnvironment xB, yBImage coordinates xO, yO, zOObject coordinates VST1-VST6Preliminary steps ST1-ST5Main steps

Claims

1. Method for determining an articulation angle (KW) between a towing vehicle (1a) and a trailer (1b) coupled thereto, wherein the trailer (1b) can be imaged in single images (EBk) having single-image pixels (EBkPi) using a camera system (2), comprising at least the following steps: - determining a point cloud (PW) from at least one single image (EBk) in which at least the trailer (1b) is imaged, wherein the point cloud (PW) consists of object points (POn) which are assigned to the trailer (1b) (VST3); - forming at least one planar reference surface (RF) from the adjacent object points (POn) of the point cloud (PW) which lie in the same plane (E) (VST4), wherein each planar reference surface (RF) lies in a fixed spatial orientation (RFO) and / or a fixed spatial position (RFP) relative to the trailer (1b); - reading in at least one single image (EBk) (ST2) in which the trailer (1b) coupled to the towing vehicle (1a) is imaged, and determining at least one partial surface (FT) in this at least one read-in single image (EBk) (ST3); characterized by - fitting the at least one previously formed planar reference surface (RF) into the at least one determined partial surface (FT) of the read-in single image (EBk) by applying a geometric transformation (TG) (ST4); - determining the articulation angle (KW) between the towing vehicle (1a) and the trailer (1b) as a function of a rotational component (TGR) of the applied geometric transformation (TG), wherein the rotational component (TGR) of the applied geometric transformation (TG) characterizes a rotation of the planar reference surface (RF) about a pivot axis (H) of the trailer (1b).

2. Method according to claim 1, characterized in that to determine the point cloud (PW) (VST3), the at least one single image (EBk) is captured when the trailer (1b) is not yet coupled to the towing vehicle (1a), for example during a coupling process (AV).

3. Method according to claim 1 or 2, characterized in that the point cloud (PW) determined from the at least one single image (EBk) and / or the at least one formed planar reference surface (RF) is stored, preferably together with the spatial orientation (RFO) and / or the spatial position (RFP) of each planar reference surface (RF) relative to the trailer (1b) (VST6).

4. Method according to any of the preceding claims, characterized in that the point cloud (PW) is determined using an SfM algorithm (AS) from at least two single images (EBk), the trailer (1b) being captured in the at least two single images (EBk) from at least two different viewpoints (SP1, SP2), preferably by a monocular camera (2a) of the camera system (2).

5. Method according to claim 3, characterized in that the point cloud (PW) is formed from a plurality of object points (POn) which are assigned to at least the trailer (1b), depth information (Tln) regarding the respective object points (POn) being obtained from the at least two single images (EBk) by triangulation (T) (VST2).

6. Method according to any of the preceding claims, characterized in that the at least one planar reference surface (RF) is defined from adjacent object points (POn) lying in the same plane (E) in such a way that the planar reference surface (RF) can be assigned to a region which is delimited or can be delimited by color and / or by structure and comprises adjacent single-image pixels (EBkPi) in the at least one single image (EBk).

7. Method according to claim 6, characterized in that the planar reference surface (RF) is formed from the adjacent object points (POn) of the point cloud (PW) which lie in the same plane (E) and / or to which single-image pixels (EBkPi) having the same or at least similar color values (CW) and / or intensity values (IW) are assigned in the at least one single image (EBk).

8. Method according to any of the preceding claims, characterized in that the at least one planar reference surface (RF) completely covers a trailer front side (1c) or only partially occupies the trailer front side (1c).

9. Method according to any of the preceding claims, characterized in that adjacent single-image pixels (EBkPi) having the same or at least similar color values (CW) and / or intensity values (IW) are assigned to the same partial surface (FT) of the read-in single image (EBk) (ST3).

10. Method according to any of the preceding claims, characterized in that - an intermediate surface (RFZ) is formed from the at least one previously formed planar reference surface (RF) by a coordinate transformation (TK), the planar reference surface (RF) being present in a three-dimensional coordinate system (K3D) and the intermediate surface (RFZ) being present in a two-dimensional coordinate system (K2D), and - the intermediate surface (RFZ) formed in each case is fitted into the relevant partial surface (TF) of the read-in single image (EBk) by the geometric transformation (TG) (ST3).

11. Method according to claim 10, characterized in that the intermediate surface (RFZ) formed in each case is shifted and / or rotated and / or sheared and / or scaled by the geometric transformation (TG), for example in a plurality of iteration steps (STI).

12. Method according to claim 10 or 11, characterized in that pixel-by-pixel color values (CW) and / or intensity values (IW) in the relevant partial surface (FT) and in the formed intermediate surface (RFZ) are compared in order to fit the at least one previously formed planar reference surface (RF) into the at least one determined partial surface (FT).

13. Processing unit (5) for a vehicle (1) comprising a towing vehicle (1a) and a trailer (1b), in particular for carrying out a method according to any of the preceding claims, wherein the processing unit (5) is designed - to determine a point cloud (PW) from at least one single image (EBk) which is captured by a camera system (2) and in which at least the trailer (1b) is imaged, wherein the point cloud (PW) consists of object points (POn) which are assigned to the trailer (1b); - to form at least one planar reference surface (RF) from the adjacent object points (POn) of the point cloud (PW) which lie in the same plane (E), wherein each planar reference surface (RF) lies in a fixed spatial orientation (RFO) and / or a fixed spatial position (RFP) relative to the trailer (1b); - to read in at least one single image (EBk) in which the trailer (1b) coupled to the towing vehicle (1a) is imaged, and to determine at least one partial surface (FT) in this at least one read-in single image (EBk); characterized by - to fit at least one previously formed planar reference surface (RF) into the at least one determined partial surface (FT) by applying a geometric transformation (TG); and - to determine an articulation angle (KW) between the towing vehicle (1a) and the trailer (1b) as a function of a rotational component (TGR) of the applied geometric transformation (TG), wherein the rotational component (TGR) of the applied geometric transformation (TG) characterizes a rotation of the planar reference surface (RF) about a pivot axis (H) of the trailer (1b).

14. Vehicle (1) comprising a towing vehicle (1a) and a trailer (1b) and having a processing unit (5) according to claim 13 and a camera system (2), preferably having a monocular camera (2a), of which the field of view (4) is or can be aligned with the trailer (1b) in the rear region (R) of the towing vehicle (1b).