Computer-implemented method for creating test data with simulated image correspondences of a rolling shutter image sensor, computer-implemented method for testing and / or validating an image processing algorithm and control unit

Abstract

The invention relates to a computer-implemented method for creating test data (11) with simulated image correspondences (9, 13, 15) of a rolling shutter image sensor with rolling shutter effect, wherein - a digital representation of a three-dimensional scene (1) in its temporal development as well as a kinematic model (3) for the movement of the rolling shutter image sensor are provided, and wherein - a digital representation of the rolling shutter image sensor (5) with an image plane and with a readout time (Δt) r ) is specified for the row-by-row or column-by-column recording of image information; - at least at one first readout time (t1) within a first readout interval with the readout duration (Δt r) for at least one first rolling shutter image point (p1) of the image plane, at least one corresponding first rolling shutter scene point (P1) is identified in the three-dimensional scene, - at least depending on the first rolling shutter scene point (P1), a second to a second, within a second readout interval with the readout duration (Δt) r ) the corresponding rolling shutter pixel (p2) is determined at the reading time (t2), - a flux vector between the first rolling shutter image point (p1) and the second rolling shutter image point (p2) in the image plane is determined as the rolling shutter image correspondence (9) and - the rolling shutter image correspondence (9) is stored as test data (11).

Description

State of the art

[0001] When cost is a factor in the selection of digital image sensors, rolling shutter sensors are typically used instead of global shutter sensors. This is especially true for applications in the automotive industry, but also in robotics and related fields. With rolling shutter sensors, a rolling shutter effect can impair the quality of sensor images, particularly in the form of distortion. This reduced image quality, in turn, affects the quality of image processing algorithms such as visual odometry and / or scene reconstruction. This impacts the quality of both comfort and safety features, such as automatic emergency braking.

[0002] Therefore, it is necessary to improve the quality of sensor images from a rolling shutter sensor and / or to train image processing algorithms for utilizing these images. It is known that actual sensor images from real rolling shutter sensors can be used for this purpose. However, the disadvantage is that creating such sensor images is time-consuming and expensive.

[0003] In the context of this technical teaching, a rolling shutter image sensor is a sensor with line-by-line exposure. The use of such a sensor generally leads to distortion of the sensor image – hereinafter also referred to as the rolling shutter effect. This distortion arises because the image lines are not exposed simultaneously, but sequentially.

[0004] In the context of this technical teaching, a global shutter image sensor is a sensor with simultaneous exposure of all image lines. The use of such a sensor generally does not lead to any distortion of the sensor image, so that no rolling shutter effect occurs with a global shutter image sensor.

[0005] In the context of this technical teaching, an image point is defined as a point in or on a sensor surface of a sensor. The image point has a first, in particular horizontal, image coordinate and a second, in particular vertical, image coordinate.

[0006] In the context of this technical teaching, an image correspondence is at least a vector, in particular a displacement vector or a flow vector, in an image plane of the sensor; more specifically, an image correspondence is either a pair of image coordinates, for example (x1,y1) & (x2,y2) or a starting point (x,y) with a displacement vector (u,v). Furthermore, in known image processing systems, image-based image correspondences are determined and / or taken into account.

[0007] In the context of this technical teaching, a line of sight is defined as a vector from the projection center of the sensor to a point in the scene. Thus, a vector from the projection center of the sensor to a first point in the scene is a first line of sight. Furthermore, a vector from the projection center of the sensor to a second point in the scene is a second line of sight. Disclosure of the invention

[0008] The computer-implemented method with the features of claim 1 has the advantage that test data is generated purely digitally. Advantageously, no data from real scenes is required. Furthermore, it makes it possible to generate a large number of test data sets quickly and cost-effectively. A digital representation of a three-dimensional scene in its temporal development, as well as a kinematic model for the movement of the rolling shutter image sensor, are provided. Additionally, a digital representation of the rolling shutter image sensor with an image plane and a readout time for the row-wise or column-wise acquisition of image information is specified.According to the invention, at least one corresponding first rolling shutter scene point in the three-dimensional scene is identified at at least one first readout time within a first readout interval with a readout duration of at least one first rolling shutter image point in the image plane. The rolling shutter scene point is independent of the rolling shutter effect. Subsequently, according to the invention, a second rolling shutter image point corresponding to a second readout time within a second readout interval with a readout duration of at least one first rolling shutter image point is determined, at least as a function of the first rolling shutter scene point. Furthermore, a flow vector between the first rolling shutter image point and the second rolling shutter image point in the image plane is determined as a rolling shutter image correspondence, and the rolling shutter image correspondence is stored as test data, in particular as a test data set.

[0009] Preferably, the digital representation of the three-dimensional scene is determined based on a measurement. Alternatively or additionally, the digital representation of the rolling shutter image sensor is preferably determined based on a measurement. Advantageously, this allows a real-world scene to be mapped and transferred into the test file. Furthermore, it advantageously emulates a real sensor, thus tailoring the test file to the actual sensor.

[0010] According to a preferred embodiment of the invention, it is provided that the second readout time is first determined, at least as a function of the first rolling shutter scene point, and that the second rolling shutter pixel is then determined, at least as a function of the first rolling shutter scene point and the second readout time. Advantageously, this enables a simple, fast, and robust determination of the second readout time and the second rolling shutter pixel.

[0011] It is particularly preferred that, using a time-scene point function, the first scene point is calculated as a second scene point at the second time point in the three-dimensional scene, at least as a function of a first scene point, a first time point, and a second time point. This thus determines the movement of the first scene point over time, from the first time point to the second time point. Advantageously, the second scene point is calculated simply and quickly using the time-scene point function. In particular, the first rolling shutter scene point is used as the first scene point. Alternatively or additionally, the first readout time is advantageously used as the first time point. Alternatively or additionally, the second readout time is advantageously used as the second time point.Alternatively or additionally, the first rolling shutter scene point is advantageously used as the first scene point. Alternatively or additionally, a second rolling shutter scene point is advantageously calculated as the second scene point.

[0012] According to a preferred embodiment of the invention, a scene-ray function is used to calculate a ray from the digital rolling shutter sensor to the scene point, at least as a function of a scene point. Advantageously, the scene-ray function calculates the ray simply and quickly. In particular, the second rolling shutter scene point is used as the scene point. Alternatively or additionally, a second ray is calculated as the ray.

[0013] It is particularly preferred that the second pixel be calculated using a light-ray-pixel function, at least as a function of the second viewing ray. Advantageously, the second pixel is calculated simply and quickly using the light-ray-pixel function. In particular, the second pixel is calculated as the second rolling shutter pixel.

[0014] According to a preferred embodiment of the invention, a relative readout time of the pixel is calculated using a pixel-time function, at least as a function of one pixel. Advantageously, the relative readout time of the pixel is calculated simply and quickly using the pixel-time function. In particular, the first rolling shutter pixel is used as the pixel, and the first relative readout time of the first rolling shutter pixel is calculated as the relative readout time of the pixel. Alternatively or additionally, the second rolling shutter pixel is used as the pixel, or the pixel is equated with the first rolling shutter pixel, and the second relative readout time of the second rolling shutter pixel is calculated as the relative readout time of the pixel.

[0015] It is particularly preferred that the second readout time is calculated using a non-linear equation, and that the non-linear equation is calculated at least as a function of the time-scene-point function, the scene-ray function, the light-ray-pixel function, the pixel-time function, a time interval between the first readout interval and the second readout interval, and a first start time of the first readout interval. t2=ts,1+Δtr+Δtw+fp(fr(f3D(fT(P1,t1,t2)))) is determined. Here, t1 is the first reading time, t2 the second reading time, t s,1 the first start time, with Δt r the readout time, with Δt W the time interval between the first readout interval and the second readout interval, with P1 being the first scene point, in particular the first rolling shutter scene point, with f r the time-scene point function, with f3D the scene view ray function, with f r the light ray pixel function and with f p The pixel-time function is denoted. Advantageously, the second readout time is determined numerically robustly using the non-linear equation. Furthermore, it is possible to easily establish the non-linear equation using the aforementioned functions.

[0016] According to a preferred embodiment of the invention, the second rolling shutter pixel is provided for by means of the equation p2=fr(f3D(fT(P1,t1,t2))) is determined. P2 is used to denote the second rolling shutter pixel.

[0017] It is particularly preferred that the first rolling shutter scene point P1 be determined using the scene line-of-sight function f 3D and the light ray pixel function f r is determined. In particular, an inverse function of the scene-ray function f is used for this purpose.3D and an inverse function of the light ray-pixel function f r used so that the first rolling shutter scene point P1 is determined using the equation P1=f3D−1(fr−1(p1)) is determined.

[0018] According to a preferred embodiment of the invention, a digital representation of a global shutter image sensor is defined along with the image plane. Furthermore, at the first start time of the first readout interval, at least one corresponding global shutter scene point in the three-dimensional scene is identified for at least one first global shutter image point of the image plane. Depending on at least the first global shutter scene point and a second start time of the second readout interval, a second global shutter image point corresponding to the first global shutter scene point is determined. Subsequently, a flow vector between the first global shutter image point and the second global shutter image point in the image plane is determined as the global shutter image correspondence. The global shutter image correspondence is stored as test data, in particular as a test data set.Advantageously, the global shutter image correspondence is created purely digitally. This also means that no data from real scenes is required. In particular, a global shutter image correspondence is assigned to a rolling shutter image correspondence.

[0019] It is particularly preferred that the second global shutter pixel be determined using the equation p2G=fr(f3D(fT(P1G,ts,1,ts,2))) is determined. This involves using t s,2 the second start time of the second readout interval, with P1G the first global shutter scene point and with p2G the second global shutter pixel. Advantageously, the same functions used for the rolling shutter sensor are employed here. In particular, the second start time of the second readout interval is determined by the equation ts,2=ts,1+Δtr+Δtw calculated.

[0020] According to a preferred embodiment of the invention, it is provided that, at least depending on the first rolling shutter scene point and the first rolling shutter image point, the image-time function and the time-scene point function, a first reference scene point can be determined by means of the equation P1R=fT(P1,−fp(p1)) is determined, where p1 is the first rolling shutter pixel and with P1R The first reference scene point is designated. Furthermore, at least depending on the first start time, the second start time, the first reference scene point, and the time-scene point function, a second reference scene point is determined using the equation P2R=fT(P1R,ts,1,ts,2) determined, whereby with P2R The second reference scene point is designated. Furthermore, at least depending on the first reference scene point, the scene ray function, and the light ray pixel function, a first reference pixel is determined using the equation p1R=fr(f3D(P1R)) determined, whereby with p1R The first reference pixel is designated. Additionally, at least depending on the second reference scene point, the scene ray function, and the light ray pixel function, a second reference pixel is determined using the equation p2R=fr(f3D(P2R)) determined, whereby with p2R The second reference pixel is designated. Subsequently, a flow vector between the first and second reference pixels in the image plane is determined as the reference image correspondence. The reference image correspondence is stored as test data. Advantageously, the reference image correspondence is created purely digitally. Advantageously, no data from real scenes is required for this. In particular, a reference image correspondence is assigned to a rolling shutter image correspondence.

[0021] The computer-implemented method according to the invention, with the features of claim 13, has the advantage that an image processing algorithm can be evaluated purely digitally, without requiring data from real scenes. Using the computer-implemented image processing algorithm, an image processing output is determined, at least as a function of a rolling shutter image correspondence of the test data generated according to the invention. The image processing output is compared with a reference image processing output, and the computer-implemented image processing algorithm is evaluated, at least as a function of this comparison.

[0022] According to a preferred embodiment of the invention, object recognition is used as the computer-implemented algorithm for image processing. The recognized object is then determined as the image processing output.

[0023] Furthermore, an existing object is used as the reference image processing output.

[0024] It is particularly preferred that a rolling shutter compensation be used as the computer-implemented algorithm for image processing. A corrected image correspondence is determined as the image processing output. Furthermore, a reference image correspondence is used as the reference image processing output.

[0025] The control unit according to the invention, with the features of claim 16, is specifically designed to execute the computer-implemented method according to the invention when used as intended. In connection with the control unit, the advantages that have already been explained in connection with the computer-implemented method for generating test data and / or the computer-implemented method for testing and / or validating a computer-implemented algorithm for image processing using test data become particularly apparent.

[0026] Further advantages and preferred features and combinations of features will become apparent in particular from the foregoing and from the claims. The invention will now be explained in more detail with reference to the drawings. To this end, we show... Fig. 1 a flowchart of an exemplary implementation of a computer-implemented method for generating test data, and Fig. 2 Flowcharts of two exemplary implementations of a computer-implemented method for testing and / or validating a computer-implemented algorithm.

[0027] Fig. Figure 1 shows a flowchart of an exemplary implementation of a computer-implemented method for creating test data.

[0028] The method requires a digital representation of a three-dimensional scene 1 in its temporal evolution and a kinematic model 3 for the movement of a rolling shutter image sensor. Additionally, a digital representation of the rolling shutter image sensor 5 with an image plane and a readout time Δt is provided. rfor the row-wise or column-wise acquisition of image information. Optionally, a digital representation of a global shutter image sensor 7 with the image plane is specified. In particular, the digital representation of the three-dimensional scene 1 and / or the digital representation of the rolling shutter image sensor 5 and / or the digital representation of the global shutter image sensor 7 is determined based on a measurement.

[0029] In step S1, at least one first readout time t1 within a first readout interval with readout duration Δt is used. r For at least one first rolling shutter image point p1 of the image plane, at least one corresponding first rolling shutter scene point P1 in the three-dimensional scene is identified. Preferably, the first rolling shutter scene point P1 is identified using a scene ray function f. 3D and a light ray pixel function f ris determined. In particular, an inverse function of the scene-ray function f is used for this purpose. 3D and an inverse function of the light ray-pixel function f r used so that the first rolling shutter scene point P1 is determined using the equation P1=f3D−1(fr−1(p1)) determined.

[0030] In step S2, at least depending on the first rolling shutter scene point P1, a second one is generated within a second readout interval with the readout duration Δt. rThe corresponding rolling shutter pixel p2 is determined for each readout time t2. In particular, the second readout time t2 is first determined, at least as a function of the first rolling shutter scene point P1. Then, the second rolling shutter pixel p2 is determined, at least as a function of both the first rolling shutter scene point P1 and the second readout time t2. Preferably, the second readout time t2 is calculated using a non-linear equation. The non-linear equation determines, at least as a function of a time-scene-point function f, the second readout time t2. T , the scene view ray function f 3D , the light ray pixel function f r , a pixel-time function f p , a time interval Δt W between the first readout interval and the second readout interval and a first start time t s,1 of the first readout interval as t2=ts,1+Δtr+Δtw+fp(fr(f3D(fT(P1,t1,t2)))) This is determined using the time-scene point function f. T The second rolling shutter scene point P2 is calculated at least as a function of the first rolling shutter scene point P1, the first readout time t1, and the second readout time t2. Furthermore, the scene line-of-sight function f is used. 3D At least depending on the second rolling shutter scene point P2, a second line of sight is calculated. Additionally, the light ray pixel function f is used. r The second rolling shutter pixel P2 is calculated, at least depending on the second line of sight. Furthermore, the pixel-time function f is used. p The second relative readout time is calculated, at least as a function of the second rolling shutter pixel p2. Subsequently, the second rolling shutter pixel p2 is determined using the equation p2=fr(f3D(fT(P1,t1,t2))) at least depending on the time-scene-point function f T , the scene view ray function f 3D and the light ray pixel function f r determined.

[0031] In step S3, a flow vector between the first rolling shutter image point p1 and the second rolling shutter image point p2 in the image plane is determined as rolling shutter image correspondence 9.

[0032] In step S4, the rolling shutter image correspondence 9 is saved as test data 11. Optionally, a global shutter image correspondence 13 and / or a reference image correspondence 15 are also saved as test data.

[0033] In an optional step S5, at the first start time t s,1 of the first readout interval for at least one first global shutter pixel p1G The image plane must contain at least one corresponding global shutter scene point. P1G identified in the three-dimensional scene.

[0034] In an optional step S6, at least depending on the first global shutter scene point, P1G and a second start time t s,2 of the second readout interval a second one, to the first global shutter scene point P1G corresponding global shutter pixel p2G determined. Preferably, the second global shutter pixel is used. p2G using the equation p2G=fr(f3D(fT(P1G,ts,1,ts,2))) at least depending on the time-scene-point function f T , the scene view ray function f 3D and the light ray pixel function f r determined.

[0035] In an optional step S7, a flow vector is created between the first global shutter pixel. p1G and the second global shutter pixel p2G in the image plane as the global shutter image correspondence 13.

[0036] In an optional step S8, at least depending on the first rolling shutter scene point P1 and the first rolling shutter pixel p1, the pixel-time function f is calculated. p and the time-scene point function f T a first reference scene point P1R using the equation P1R=fT(P1,−fp(p1)) determined.

[0037] In an optional step S9, at least depending on the first start time t s,1 , the second start time t s,2 , the first reference scene point P1R and the time-scene point function f T a second reference scene point P2R using the equation P2R=fT(P1R,ts,1,ts,2) determined.

[0038] In an optional step S10, at least depending on the first reference scene point P1R the scene view ray function f 3D and the light ray pixel function f r a first reference pixel p1R using the equation p1R=fr(f3D(P1R)) Determined. Additionally, at least depending on the second reference scene point, P2R the scene view ray function f 3D and the light ray pixel function f r a second reference pixel p2R using the equation p2R=fr(f3D(P2R)) determined.

[0039] In an optional step S11, a flow vector is created between the first reference pixel. p1R and the second reference pixel p2R in the image plane as the reference image correspondence 15.

[0040] Fig. Figure 2 shows two embodiments of a computer-implemented method for testing and / or validating a computer-implemented algorithm 17 for image processing using test data 11, each in the form of a flowchart.

[0041] In a test step T1, an image processing output 19 is determined using the computer-implemented algorithm 17 for image processing, at least as a function of a rolling shutter image correspondence 9 of the test data 11.

[0042] In a test step T2, the image processing output 19 is compared with a reference image processing output 21, resulting in a comparison 23.

[0043] In a test step T3, the computer-implemented algorithm 17 for image processing is evaluated at least in relation to the comparison 23, receiving a rating of 25.

[0044] In the first embodiment according to Fig. 2 a) A computer-implemented algorithm 17 is used for image processing, specifically object recognition 27. Additionally, a recognized object 29 is determined as an image processing output 19. Furthermore, an existing object 31 is used as a reference image processing output 21.

[0045] In the second embodiment according to Fig. 2 b) A rolling shutter compensation 33 is used as a computer-implemented algorithm 17 for image processing. Additionally, a corrected image correspondence 35 is determined as an image processing output 19. Furthermore, a reference image correspondence 15 is used as a reference image processing output 21.

Claims

[1] Computer-implemented method for creating test data (11) with simulated image correspondences (9, 13, 15) of a rolling shutter image sensor with rolling shutter effect, wherein - a digital representation of a three-dimensional scene (1) in its temporal development as well as a kinematic model (3) for the movement of the rolling shutter image sensor are provided, and wherein - a digital representation of the rolling shutter image sensor (5) with an image plane and with a readout time (Δt) r ) is specified for the row-by-row or column-by-column recording of image information; characterized by , that - at least at one first readout time (t1) within a first readout interval with the readout duration (Δt r) for at least one first rolling shutter image point (p1) of the image plane, at least one corresponding first rolling shutter scene point (P1) is identified in the three-dimensional scene, that - at least depending on the first rolling shutter scene point (P1), a second to a second, within a second readout interval with the readout duration (Δt) r ) the corresponding rolling shutter pixel (p2) at the reading time (t2) is determined, that - a flux vector between the first rolling shutter image point (p1) and the second rolling shutter image point (p2) in the image plane is determined as the rolling shutter image correspondence (9) and that - the rolling shutter image correspondence (9) is stored as test data (11). [2] Method according to claim 1, characterized by that the digital representation - the three-dimensional scene (1), and / or - of the rolling shutter image sensor (5) is determined based on a measurement. [3] Method according to any one of the preceding claims, characterized by , that - first, at least depending on the first rolling shutter scene point (P1), the second readout time (t2) is determined, and that - then, at least depending on the first rolling shutter scene point (P1) and the second readout time (t2), the second rolling shutter image point (p2) is determined. [4] Method according to any one of the preceding claims, characterized by , that by means of a time-scene point function (f r) at least depending on a first scene point, in particular the first rolling shutter scene point (P1), a first time point, in particular the first readout time (t1), and a second time point, in particular the second readout time (t2), the first scene point, in particular the first rolling shutter scene point (P1), is calculated at the second time point, in particular the second readout time (t2), in the three-dimensional scene as a second scene point, in particular a second rolling shutter scene point (P2). [5] Method according to any one of the preceding claims, characterized by , that by means of a scene ray function (f 3D ) at least depending on a scene point, in particular the second rolling shutter scene point (P2), a line of sight from the digital rolling shutter sensor to the scene point, in particular a second line of sight, is calculated. [6] Method according to any one of the preceding claims, characterized by , that by means of a light ray pixel function (f r ) at least depending on the second line of sight, the second pixel, in particular the second rolling shutter pixel (P2), is calculated. [7] Method according to any one of the preceding claims, characterized by , that by means of a pixel-time function (f p ) at least as a function of a pixel, in particular the first rolling shutter pixel (p1) and / or the second rolling shutter pixel (P2), a relative readout time of the pixel, in particular a first relative readout time of the first rolling shutter pixel (p1) and / or a second relative readout time of the second rolling shutter pixel (P2), is calculated. [8] Method according to any one of the preceding claims, characterized by, that the second readout time is calculated using a non-linear equation, and that the non-linear equation depends at least on the time-scene-point function (f T ), the scene view ray function (f 3D ), the light ray pixel function (f r ) the pixel-time function (f p ), a time interval ( Δt w ) between the first readout interval and the second readout interval and a first start time (t s,1 ) of the first readout interval as t2=ts,1+Δtr+Δtw+fp(fr(f3D(fT(P1,t1,t2)))) is determined. [9] Method according to any one of the preceding claims, characterized by , that the second rolling shutter pixel (p2) is determined by the equation p2=fr(f3D(fT(P1,t1,t2))) at least depending on the time-scene point function (f T ), the scene view ray function (f 3D ) and the light ray pixel function (fr ) is determined. [10] Method according to any one of the preceding claims, characterized by , that - a digital representation of a global shutter image sensor (7) with the image plane specified that - at the first start time (t s,1 ) of the first readout interval for at least one first global shutter pixel (p1G) The image plane must contain at least one corresponding global shutter scene point. (P1G) in the three-dimensional scene it is identified that - at least depending on the first global shutter scene point (P1G) and a second start time (t s,2 ) of the second readout interval, a second one, corresponding to the first global shutter scene point (P1G) corresponding global shutter pixel (p2G) It is determined that - a flow vector between the first global shutter pixel (p1G) and the second global shutter pixel (p2G) in the image plane is determined as global shutter image correspondence (13), and that - the global shutter image correspondence (13) is stored as test data (11). [11] Method according to any one of the preceding claims, characterized by that the second global shutter pixel (p2G) using the equation p2G=fr(f3D(fT(P1G,ts,1,ts,2))) at least depending on the time-scene point function (f T ), the scene view ray function (f 3D ) and the light ray pixel function (f r ) is determined. [12] Method according to any one of the preceding claims, characterized by , that - at least depending on the first rolling shutter scene point (P1) and the first rolling shutter pixel (p1), the pixel-time function (f p ) and the time-scene point function (f T ) a first reference scene point (P1R) using the equation P1R=fT(P1,−fp(p1)) It is determined that - at least depending on the first start time (t s,1 ), the second start time (t s,2 ), the first reference scene point (P1R) and the time-scene point function (f T ) a second reference scene point P2R using the equation P2R=fT(P1R,ts,1,ts,2) It is determined that - at least depending on the first reference scene point (P1R), the scene view ray function (f 3D ) and the light ray pixel function (f r ) a first reference pixel (p1R) using the equation p1R=fr(f3D(P1R)) It is determined that - at least depending on the second reference scene point (P2R), the scene view ray function (f 3D ) and the light ray pixel function (f r ) a second reference pixel (p2R) using the equation p2R=fr(f3D(P2R)) It is determined that - a flow vector between the first reference pixel (p1R) and the second reference pixel (p2R) in the image plane is determined as the reference image correspondence (15), and that - the reference image correspondence (15) is stored as test data (11). [13] Computer-implemented method for testing and / or validating a computer-implemented algorithm (17) for image processing using test data (11) according to any of the preceding claims, characterized by , that - by means of the computer-implemented algorithm (17) for image processing, at least depending on a rolling shutter image correspondence (9) of the test data (11), an image processing output (19) is determined, that - the image processing output (19) is compared with a reference image processing output (21), and that - the computer-implemented algorithm (17) for image processing is evaluated at least in relation to the comparison (23). [14] Method according to claim 12, characterized by, that an object recognition algorithm (27) is used as the computer-implemented algorithm (17) for image processing, that a recognized object (29) is determined as the image processing output (19), and that an existing object (31) is used as the reference image processing output (21). [15] Method according to claim 12, characterized by , that a rolling shutter compensation (33) is used as the computer-implemented algorithm (17) for image processing, that a corrected image correspondence (35) is determined as the image processing output (19), and that a reference image correspondence (15) is used as the reference image processing output (21). [16] Control unit specifically designed to perform a computer-implemented method according to any of the preceding claims when used as intended.

Images