Road marking detection

Abstract

According to the method for road marking detection, sensor data sets depicting a road marking (7) at a first and a second measurement instant are generated by an environmental sensor system (4), and parameters characterizing a motion of the environmental sensor system (4) are determined. First and second observation state vectors describing the road marking (17) at the first and the second measurement instant, respectively, are generated on the basis of the sensor data sets. A predicted state vector for the second measurement instant is calculated from the at least one motion parameter and the first observation state vector, and a corrected state vector for the second measurement instant is generated from the predicted state vector and the second observation state vector.

Description

Technical Field

[0001] This invention relates to a method for road marking detection, wherein a first sensor dataset depicting road markings at a first measurement time and a second sensor dataset depicting road markings at a second measurement time are generated using an environmental sensor system. The invention also relates to a method for at least partially automated guidance of motor vehicles, a road marking detection system, an electronic vehicle guidance system, a computer program, and a computer-readable storage medium. Background Technology

[0002] Advanced Driver Assistance Systems (ADAS), and systems used for automatically or autonomously guiding motor vehicles, utilize data generated by environmental sensor systems (such as cameras or lidar systems) installed at various locations on the vehicle. In particular, data generated by environmental sensor systems can be used to identify and detect road markings on the road the vehicle is traveling on. The location, shape, and type of road markings are important inputs for automated or semi-automated driving functions. However, external interference or noise can impair the reliability of road marking detection, and thus the reliability of subsequent functions using the road marking detection results.

[0003] Document US7933433B2 describes a device for recognizing lane markings. This device is capable of extracting data from lane markings based on the position, shape, and width of the centerline of the lane projected for the current cycle. Summary of the Invention

[0004] Therefore, the object of the present invention is to provide an improved concept for road marking detection that increases reliability.

[0005] This objective is achieved through the relevant subject matter of the independent claims. Further embodiments and preferred embodiments are the subject matter of the dependent claims.

[0006] Based on an improved concept, a method for road marking detection is provided. This method involves generating a first sensor dataset depicting road markings, particularly those located on the road, at a first measurement time, and a second sensor dataset depicting the road markings at a second measurement time, using an environmental sensor system, particularly an environmental sensor system of a motor vehicle traveling on a road. At least one motion parameter characterizing the motion of the environmental sensor system relative to the road markings between the first and second measurement times is determined. A computational unit generates a first observation state vector based on the first sensor dataset, wherein the first observation state vector geometrically describes the road markings at the first measurement time. The computational unit also generates a second observation state vector geometrically describing the road markings at the second measurement time based on the second sensor dataset. Furthermore, the computational unit calculates a predicted state vector at the second measurement time based on at least the motion parameter and the first observation state vector. Finally, the computational unit generates a corrected state vector at the second measurement time based on the predicted state vector and the second observation state vector.

[0007] Road markings, also known as lane markings or pavement markings, can include one or more straight lines or curves, such as single lines, double lines, triple lines, etc. The lines can be continuous or can include discontinuous segments. Similarly, road markings can include combinations of lines of the same type and / or different types.

[0008] An environmental sensor system can be understood as a sensor system capable of generating sensor data or sensor signals that depict, represent, or image the environment in which the environmental sensor system is located. For example, a camera, lidar system, radar system, or ultrasonic sensor system can be considered an environmental sensor system. Preferably, an environmental sensor system includes a camera system or a lidar system.

[0009] Environmental sensor systems can generate corresponding sensor datasets that correspond to a time series of consecutive discrete measurement moments, including first and second measurement moments. Each measurement moment can correspond to a discrete point in time. However, depending on the type of environmental sensor system used, measurements at a given measurement moment may require a certain time interval. For example, in the case of a camera as an environmental sensor system, each measurement moment may correspond to a corresponding video frame. In the case of a scanning lidar system, each measurement moment may correspond to a scanning cycle, etc.

[0010] The second measurement time follows the first measurement time, and in particular, directly follows the first measurement time. Typically, the environmental sensor system and the vehicle move between the first and second measurement times, causing the relative position and / or orientation of the road markings with respect to the environmental sensor system to change between the first and second measurement times.

[0011] To generate the corresponding observation state vector, the computing unit can determine a geometric approximation of the road markings based on the corresponding sensor dataset, and store two or more parameters as the corresponding observation state vector according to this approximation. The geometric approximation of the road markings can, for example, be performed in a predetermined coordinate system rigidly connected to the sensor system.

[0012] For example, a coordinate system can be defined by a longitudinal axis, a transverse axis, and a normal axis, wherein all three axes are perpendicular to each other. The longitudinal and transverse axes can, for example, define planes that are generally parallel to the road surface. The longitudinal, transverse, and normal axes can also correspond to the respective axes of a motor vehicle, or can be transformed to these axes by a predetermined coordinate transformation, which can be determined, for example, by sensor calibration.

[0013] To approximate road markings in a predetermined coordinate system, the computational unit can, for example, perform a polynomial approximation. For instance, road markings can be approximated using a quadratic polynomial. In this case, the coefficients of the corresponding polynomial can correspond to the constant position or offset of the road marking or its defined point relative to one of the axes of the coordinate system (e.g., the longitudinal axis). Other coefficients of the corresponding polynomial can correspond to the orientation and curvature of the road marking at the corresponding point.

[0014] The observed state vector may include, for example, polynomial coefficients obtained from approximations or quantities derived from polynomial coefficients.

[0015] At least one motion parameter may be determined, for example, by one or more motion sensors of the vehicle, such as an acceleration sensor, a yaw rate sensor, etc., or may depend on the output of the motion sensor. For example, at least one motion parameter may include the longitudinal velocity of the environmental sensor system in the longitudinal axis direction and / or the lateral velocity of the environmental sensor system in the transverse axis direction and / or the yaw rate of the environmental sensor system relative to the normal axis.

[0016] A predicted state vector is calculated based on a first observed state vector, which undergoes state dynamics defined by at least one motion parameter. Then, the predicted state vector is corrected based on the actual observation at a second measurement moment, i.e., based on a second observed state vector. This reduces errors caused by external disturbances or noise that inevitably affect the observed state vector. Therefore, compared to the second observed state vector, the reliability of the corrected state vector for subsequent applications, such as driver assistance functions or autonomous driving functions, is improved.

[0017] At least one motion parameter may be affected by noise, which could reduce the predictive quality of the state vector. Nevertheless, the eventual availability of the predicted state allows for the correction of the state vector based on the actual observation at a second measurement moment, i.e., based on the second observed state vector. Therefore, predictions can be made even if at least one motion parameter is not measurable at all, for example, by assuming a historical value or estimate of at least one motion parameter.

[0018] Specifically, the corrected state vector can be considered as part of the output of road mark detection. The method steps described for the first and second measurement times can be performed iteratively or repeatedly for multiple consecutive measurement times to obtain a time series of the corrected state vector that contributes to the output of road mark detection. In this way, a particularly reliable input is provided for autonomous driving functions or driver assistance functions.

[0019] By representing road markings using corresponding parameters that geometrically describe their underlying patterns, the output of road marking detection can be stabilized. In particular, representing road markings using state vectors allows for reliable prediction and correction compared to predictions based on sensor data itself (e.g., points in a point cloud). In other words, the state of road markings can be effectively tracked instead of traditional object tracking.

[0020] According to various embodiments of the method for road mark detection, a computing unit is used to generate two or more first sampled state vectors at a first measurement time based on a first observed state vector and at least one parameter describing a multidimensional distribution. A predicted state vector is then calculated based on the two or more first sampled state vectors.

[0021] At least one parameter describing the multidimensional distribution may, for example, include the corresponding variance and / or covariance of the quantities contained in the input of the first observation state vector. For example, the computational unit may determine the covariance matrix for each measurement moment, particularly the first measurement moment. For instance, representing the first observation state vector as x0 = (a, b, …), the corresponding covariance matrix is:

[0022]

[0023] Among them, the variance contained in the covariance matrix ², b ², …and covariance ab , ba …can be determined, for example, by a computing unit based on a sensor dataset generated by the environmental sensor system for measurement times prior to the first measurement time. Alternatively or additionally, variance… ², b², …and covariance ab , ba The values ​​of , … can be output by the environmental sensor system if these values ​​are known in advance, for example, through calibration. A combination of the two methods is also possible.

[0024] Then, two or more first sampled state vectors can be parameterized, for example, as follows:

[0025]

[0026] in, is the scaling parameter, n is the dimension of the state vector, and α and κ are predefined parameters for the diffusion of control points relative to the mean µ, which can be given, for example, by the input of x0. The subscript i selects the i-th column vector of the covariance matrix S.

[0027] Thus, the first observed state vector, together with two or more first sampled state vectors at the first measurement time, forms a population of corresponding distributions described by at least one parameter α, κ that describes the multidimensional distribution. To compute the predicted state vector at the second measurement time, the population at the first measurement time, including the first observed state vector and two or more first sampled state vectors, can be projected onto the second measurement time according to at least one motion parameter.

[0028] This allows for an effective way to minimize the impact of errors or disturbances on the sensor dataset. In particular, the actual distribution of the inputs on the observed state vector does not necessarily follow a Gaussian distribution.

[0029] According to several implementations, the computing unit is used to determine a first polynomial based on a first sensor dataset to approximate road markings at a first time, and to determine a second polynomial based on the sensor dataset to approximate road markings at a second time. A first observation state vector includes the coefficients of the first polynomial, specifically all the coefficients of the first polynomial, and a second observation state vector includes the coefficients of the second polynomial, specifically all the coefficients of the second polynomial.

[0030] The first and second polynomials are, for example, first-order or higher-order polynomials, preferably second-order polynomials. In this case, the coefficients of the first and second polynomials can be understood to relate to constant offset, direction, and curvature, respectively.

[0031] The computational unit can execute a fitting algorithm to determine first and second polynomials based on the first and sensor datasets, respectively. In this way, a fairly accurate estimate of road markings can be obtained without involving computationally expensive methods. In particular, representing road markings with polynomial coefficients is more efficient than the raw output of the tracking environment sensor system.

[0032] According to several embodiments, the calculation unit is used to calculate the prediction covariance matrix at the second measurement time based on at least one motion parameter, a first observation state vector, and a prediction state vector. The calculation unit is also used to generate a correction covariance matrix at the second measurement time based on the prediction covariance matrix and the second observation state vector.

[0033] The output of road mark detection may include a corrected covariance matrix, and, if applicable, may also include a corresponding corrected covariance matrix for other measurement times.

[0034] The corrected covariance matrix provides a suitable means to assess the spread and distribution of parameters describing road markings, and thus the reliability of measurements used for driver assistance or autonomous driving functions.

[0035] As described regarding the predicted state vector at the second time step, the predicted covariance matrix at the second measurement time step can also be calculated based on two or more first sampled state vectors.

[0036] According to several embodiments, an environmental sensor system is used to generate multiple sensor datasets depicting road markings at corresponding consecutive measurement times, wherein the multiple sensor datasets include a first sensor dataset and a second sensor dataset. For each measurement time, at least one motion parameter characterizing the motion of the environmental sensor system relative to the road markings between the corresponding measurement time and a corresponding subsequent measurement time is determined. For each measurement time, a computing unit is used to generate a corresponding observed state vector that geometrically describes the road markings at the corresponding measurement time based on the corresponding sensor dataset. For each measurement time, or for each measurement time except the final measurement time, the computing unit is used to calculate a predicted state vector for the corresponding subsequent measurement time based on the corresponding at least one motion parameter between the corresponding measurement time and the corresponding subsequent measurement time, and based on the observed state vector at the corresponding measurement time. For each measurement time, or for each measurement time except the final measurement time, the computing unit is used to generate a corrected state vector for the corresponding subsequent measurement time based on the predicted state vector and the observed state vector at the corresponding subsequent measurement time.

[0037] The interpretation of the first and second observation state vectors, as well as the first and second measurement times, can be similarly applied to each measurement time and its corresponding subsequent measurement time in a plurality of consecutive measurement times. Specifically, each measurement time, except for the final measurement time, has exactly one corresponding subsequent measurement time. Therefore, as a result of the method for road mark detection, a time series of the corrected state vectors is obtained.

[0038] Similarly, in a further embodiment, a time series of the corrected covariance matrix can also be obtained.

[0039] According to the improved concept, a method for at least partially automatically guiding a motor vehicle is also provided. For this purpose, a method for road marking detection according to the improved concept is implemented, wherein an environmental sensor system is mounted on the motor vehicle. Based on a corrected state vector, and particularly by means of the vehicle's electronic vehicle guidance system, the motor vehicle is guided at least partially automatically.

[0040] In particular, the computing unit is composed of a motor vehicle, for example, an electronic vehicle guidance system.

[0041] Electronic vehicle guidance systems can be understood as electronic systems configured to guide vehicles in a fully automatic or fully autonomous manner, and specifically, without requiring manual intervention or control by the vehicle's driver or user. The vehicle performs all necessary functions, such as steering, deceleration, and / or acceleration, as well as automatically monitoring and recording road traffic and corresponding reactions. Specifically, according to Level 5 of SAE J3016 classification, electronic vehicle guidance systems can achieve fully automatic or fully autonomous driving modes. Electronic vehicle guidance systems can also be implemented as Advanced Driver Assistance Systems (ADAS), assisting the driver in partially automatic or partially autonomous driving. Specifically, according to Levels 1 to 4 of SAE J3016 classification, electronic vehicle guidance systems can achieve partially automatic or partially autonomous driving modes. Here and below, SAE J3016 refers to the corresponding standard dated June 2018.

[0042] Therefore, at least partially automated vehicle guidance may include guiding the vehicle using a fully automated or fully autonomous driving mode at Level 5 according to SAE J3016 classification. At least partially automated vehicle guidance may also include guiding the vehicle using a partially automated or partially autonomous driving mode at Levels 1 to 4 according to SAE J3016 classification.

[0043] At least partially automated guidance of motor vehicles may include, for example, lateral control of the motor vehicle.

[0044] According to several embodiments of a method for at least partially automatically guiding motor vehicles, a computing unit is used to generate electronic map data representing the motor vehicle's environment based on a correction state vector.

[0045] According to several implementations, a motor vehicle is guided at least partially automatically based on electronic map data, or another motor vehicle is guided at least partially automatically based on electronic map data.

[0046] This provides a particularly reliable way to generate electronic map data (such as high-definition maps, HD maps).

[0047] According to various implementations, particularly by using the control unit of an electronic vehicle guidance system, one or more control signals for guiding motor vehicles are generated at least partially automatically based on a correction state vector or electronic map data.

[0048] One or more control signals may be provided, for example, to one or more corresponding actuators to control the movement of the motor vehicle.

[0049] According to the improved concept, a road marking detection system is also provided. The road marking detection system includes an environmental sensor system configured to generate a first sensor dataset depicting road markings at a first measurement time and a second sensor dataset depicting road markings at a second measurement time. The road marking detection system includes at least one motion sensor system configured to determine at least one motion parameter characterizing the motion of the environmental sensor system relative to the road markings between the first and second measurement times. The road marking detection system includes a computing unit configured to generate a first observation state vector geometrically describing the road markings at the first measurement time based on a first predefined segment, and to generate a second observation state vector geometrically describing the road markings at the second measurement time based on the sensor dataset. The computing unit is configured to compute a predicted state vector at the second measurement time based on at least one motion parameter and the first observation state vector, and to generate a corrected state vector at the second measurement time based on the predicted state vector and the second observation state vector.

[0050] Based on the improved concept, and according to multiple implementations of the road marking detection system, the environmental sensor system includes a lidar system.

[0051] In such an implementation, the first and second sensor datasets may include corresponding point clouds generated from sensor signals from a lidar system.

[0052] Other embodiments of the road marking detection system are directly derived from various embodiments of the road marking detection method according to the improved concept and the method for at least partially automatically guiding motor vehicles according to the improved concept, and vice versa. In particular, the road marking detection system can be configured to perform the method according to the improved concept or perform the method according to the improved concept.

[0053] According to the improved concept, an electronic vehicle guidance system including a road marking detection system according to the improved concept is also provided. The electronic vehicle guidance system includes a control unit configured to at least partially automatically generate one or more control signals for guiding a motor vehicle based on a correction state vector.

[0054] The computing unit may include a control unit, or the control unit and the computing unit may be implemented separately.

[0055] According to the improved concept, a motor vehicle is also provided, which includes an electronic vehicle guidance system or a road marking detection system according to the improved concept.

[0056] According to the improved concept, a first computer program including first instructions is also provided. When the road marking system executes the first instructions or the first computer program according to the improved concept, the first instructions cause the road marking system to perform a method for road marking detection according to the improved concept.

[0057] According to the improved concept, a second computer program including the second instructions is also provided. When the second instructions or the second computer program are executed by the electronic vehicle guidance system according to the improved concept, the second instructions cause the electronic vehicle guidance system to perform a method for at least partially automatically guiding a motor vehicle according to the improved concept.

[0058] According to the improved concept, a computer-readable storage medium is also provided. The computer-readable storage medium stores a first computer program according to the improved concept and / or a second computer program according to the improved concept.

[0059] Further features of the invention will be apparent from the claims, drawings, and description of the drawings. Features and combinations of features mentioned in the foregoing description, as well as features and combinations of features mentioned and / or shown individually in the drawings below, can be included not only in the improved concept within the separately specified combinations, but also in other combinations. Therefore, embodiments of the improved concept are included and disclosed, which may not be explicitly shown or explained in the drawings, but are generated and produced by separate combinations of features from the explained embodiments. Embodiments and combinations of features that do not have all the features of the initially stated claims can be included in the improved concept. Furthermore, embodiments and combinations of features that extend beyond or deviate from the combinations of features set forth in the claims can be included in the improved concept. Attached Figure Description

[0060] In the attached diagram:

[0061] Figure 1 A motor vehicle including an electronic vehicle guidance system according to an improved concept is schematically shown;

[0062] Figure 2 An exemplary implementation of an environmental sensor system is illustrated schematically;

[0063] Figure 3 Another exemplary implementation of the environmental sensor system is illustrated schematically;

[0064] Figure 4 An example of a lidar point cloud is shown schematically;

[0065] Figure 5 This schematically illustrates the possibility of geometrically approximating road markings;

[0066] Figure 6 An exemplary time series of state parameters for road markings is shown;

[0067] Figure 7 It shows Figure 6 The distribution of parameters;

[0068] Figure 8 Another exemplary time series of state parameters for road markings is shown; and

[0069] Figure 9 It shows Figure 8 The distribution of parameters. Detailed Implementation

[0070] Figure 1 A motor vehicle 1 equipped with an electronic vehicle guidance system 3 is shown for at least partially automatic guidance of the motor vehicle 1. The electronic vehicle guidance system 3 includes a road marking detection system 2 according to an improved concept.

[0071] The road marking detection system 2 includes an environmental sensor system 4, which is implemented, for example, as a lidar system. The road marking detection system 2 also includes a computing unit 5 connected to the environmental sensor system 4 and one or more motion sensors 6 connected to the computing unit 5. The motion sensors 6 may include one or more acceleration sensors and / or yaw rate sensors, etc.

[0072] Figure 2 The construction of an exemplary embodiment of an environmental sensor system 4, serving as a lidar laser scanner, is schematically shown. Figure 2 The left side depicts the launch phase, while the right side depicts the testing phase.

[0073] The environmental sensor system 4 includes an emitting unit 7, such as one or more infrared laser sources, configured to emit pulsed light 9 into the environment of the motor vehicle 1. For this purpose, a modulation unit 8 (e.g., a rotatable mirror) is arranged to deflect the pulsed light 9 in different horizontal directions depending on the position of the modulation unit 8. The initially collimated laser can be defocused to produce a beam with a non-zero azimuth angle, such that the pulsed light 9 covers a solid angle range 10 in the environment.

[0074] The environmental sensor system 4 also includes a detection unit 11, which contains, for example, a set of photodetectors or a photodetector array, capable of collecting the reflected portion of the pulsed light 9 incident from a corresponding portion of the solid angle range 10, and converting the detected reflected light into a corresponding detector signal. The computing unit 5 can evaluate the detector signal to determine the corresponding location of the reflection point in the environment and the distance of that point from the environmental sensor system 4, for example, through time-of-flight measurements.

[0075] As a result, it provided, such as Figure 4 The point cloud 16, schematically depicted, is composed of measurements output in a spherical coordinate system. The time-of-flight measurement produces the corresponding radial distance, the angular position of the modulation unit 8 determines the horizontal or azimuth angle, and the position of the corresponding optical detector 12 relative to the modulation unit 8 determines the polar or vertical angle. Figure 4 An example of the corresponding point cloud 16 projected onto a two-dimensional plane parallel to the road surface is shown.

[0076] like Figure 3 As shown, the environmental sensor system 4 installed on the motor vehicle 1 traveling on the road 14 can generate multiple layers of scanning points, each corresponding to a specific optical detector 12. Depending on whether the corresponding light is reflected from the surface of the road 14, these layers can be represented as a ground layer 15 or a non-ground layer 20, respectively. Thus, the ground layer 15 can depict or represent road markings 17 on the road surface of the road 14, such as... Figure 4 As shown.

[0077] The computing unit 5 can, for example, be made by means of, such as Figure 5 The quadratic polynomial 17' shown is used to approximate the road marking 17 depicted in the point cloud 16. Using the quadratic polynomial 17', the road marking 17 can be approximately described as:

[0078]

[0079]

[0080]

[0081]

[0082] Where x and y correspond to the longitudinal and lateral coordinates of the sensor coordinate system rigidly connected to the environmental sensor system 4, respectively, as shown below. Figure 4As shown. d represents a constant lateral offset, for example, at x = 0; θr represents an approximate angle, where the tangent to the quadratic polynomial 17′ makes an angle with the longitudinal axis at x = 0; and r represents the approximate radius of curvature at x = 0. Therefore, the calculation unit 5 can generate the observation state vector for the corresponding measurement time given by x0 = (a, b, c). Furthermore, the environmental sensor system can output information that allows the calculation unit 5 to determine the covariance matrix for the corresponding measurement time:

[0083]

[0084] Alternatively, the computational unit can determine the covariance matrix based on previously measured distributions and corresponding state vectors.

[0085] Thus, the calculation unit 5 generates an observation state vector and a corresponding covariance matrix for each of multiple consecutive measurement moments. Furthermore, the calculation unit 5 continuously or repeatedly reads the output of the motion sensor 6 to determine the longitudinal velocity v of the motor vehicle 1. x Lateral velocity v y And the corresponding value of the yaw rate ω, and thus determine the corresponding value of the environmental sensor system 4.

[0086] Computational unit 5 applies a nonlinear state estimation algorithm to predict the state or condition of each road mark 17 relative to the vehicle 1. Specifically, it does not require inferring the distances between different road marks 17. Compared to standard linear estimators such as the standard Kalman filter, this method has the advantage of not being limited to linear systems or Gaussian distributions. Therefore, better accuracy is achieved without increasing computational complexity. The state estimation algorithm consists of two main stages: time prediction and time correction.

[0087] The computing unit can generate multiple sampled state vectors based on the corresponding state vector at each measurement time according to the rules:

[0088]

[0089] in, is the scaling parameter, n is the dimension of the state vector, and α and κ are predefined parameters for the diffusion of control points relative to the mean µ, which can be given, for example, by the input of x0. The subscript i selects the i-th column vector of the covariance matrix S.

[0090] During the time projection stage, computation unit 5 generates the corresponding predicted state vector by projecting the observed state vector and sampled state vector at a given measurement time k-1 to the subsequent measurement time k based on the fundamental dynamics of the system represented by the state transition function.

[0091] ,

[0092] in, It represents the time between two consecutive measurement moments.

[0093] The result of the time projection stage is given by the following formula:

[0094]

[0095]

[0096]

[0097]

[0098]

[0099] The first equation estimates the state vector and thus projects the variable to be tracked from the previous measurement time to the current measurement time. The second equation projects the covariance matrix of the state vector to the current measurement time. The third and fourth equations project the state vector into the measurement space to fit it to an appropriate dimension. The fifth equation calculates the covariance of the measurements. In this simple example, the state vector already has the dimension of the measurement space, so h represents the units. Furthermore, Describes x including x0 and measurement time k-1. i As a matrix of columns, summation is performed on i, and for each addend, f is applied to the matrix. The corresponding column i. and This represents the weights used to prioritize samples. and This represents the uncertainty that the model cannot describe and the sensor representation cannot cover.

[0100] During the time correction phase, taking into account the difference between the system gain-weighted projection estimate and the actual measurement, the state vector and its covariance matrix are updated:

[0101]

[0102]

[0103]

[0104] In this process, the third equation calculates the system gain, while the first and second equations consider updating the state vector and its covariance by the difference between the gain-weighted estimate and the measurement. These updates can be considered as representing the output of the estimate, representing the output of the nonlinear state estimator.

[0105] As a result, time series of corrected or updated state vectors and covariance matrices are obtained. The described method independently interprets each road sign polynomial representation given by the corresponding state vector as a distribution, rather than just a unique sample.

[0106] By means of the described mechanism, particularly the combination of polynomial representation and polynomial state estimation, the output of this method includes a polynomial representation of each road mark 17, which exhibits reduced variation in lateral offset, orientation, and curvature, and in particular a smaller standard deviation. Furthermore, due to the time projection step, it is able to provide a certain amount of continuity to the signal even when it cannot be measured or when the amount of noise present makes the measurement unreliable.

[0107] The above effects can be observed under different driving conditions. For example, in Figure 6 In the driving operation, the input of the state vector is given as a function of time, where motor vehicle 1 is traveling on straight road 14 and performing lane changes and returns. Figure 6 The top figure shows the curvature 1 / (2r), the middle figure shows the direction tan(θ), and the bottom figure shows the lateral offset d. Figure 6 The test results for the polynomial captured as an empty symbol over time are shown. The deviation is quite large. Solid symbols represent the output of the method based on the improved concept. The resulting curve is sharper, and the deviation is significantly reduced.

[0108] Figure 7 The figure above shows the corresponding distribution 18 of the original values ​​and the distribution 18' of the output of the nonlinear state estimator with curvature 1 / (2r). The figure below shows the corresponding distributions 19 and 19' of the direction tan(θ).

[0109] exist Figure 8 In, similar to Figure 6 The symbol is now shown for a more acute situation, namely driving at a highway exit where it becomes a sharp turn. Figure 8 It demonstrates how to track detected changes, particularly changes in curvature, but with significantly reduced noise through a method based on an improved concept. Furthermore, it can track sudden changes in direction very well.

[0110] Figure 9 It shows that for Figure 8 The upper figure shows the curvature distribution 18, 18', and the lower figure shows the direction distribution 19, 19'. Here, it can be clearly seen that the method also works well for non-Gaussian distributions.

[0111] As described, and particularly with respect to the accompanying drawings, the improved concept allows for stable road mark detection that closely follows actual road marks in a consistent and reliable manner. Road marks are represented by a corresponding set of parameters describing their basic pattern and can be applied to a nonlinear state estimator for stable detection relative to parameter variations between cycles.

[0112] The improved concept can handle driving maneuvers such as lane changes and sharp turns. It can reduce or even avoid sudden changes in the detected time series. In implementations using nonlinear state estimators, non-Gaussian processes can be handled without needing to be linearized. Furthermore, the computational cost is not increased, or at least not significantly increased, compared to linear state estimators.

Claims

1. A method for road marking detection, wherein a first sensor dataset depicting road markings (17) at a first measurement time and a second sensor dataset depicting road markings (17) at a second measurement time are generated by using an environmental sensor system (4), said environmental sensor system (4) including a lidar system; Its features are, - Determine at least one motion parameter characterizing the motion of the environmental sensor system (4) relative to the road markings (17) between the first and second measurement times; and - The calculation unit (5) is used for: - Based on the first sensor dataset, a first observation state vector is generated to geometrically describe the road marking (17) at the first measurement time. The first sensor dataset includes a point cloud generated based on the sensor signals of the lidar system. The computing unit (5) performs a fitting algorithm to determine a first polynomial based on the first sensor dataset to approximate the road marking (17) at the first measurement time. The first observation state vector includes the coefficients of the first polynomial. - Based on the second sensor dataset, a second observation state vector is generated to geometrically describe the road marking (17) at the second measurement time. The second sensor dataset includes a point cloud generated based on the sensor signals of the lidar system. The computing unit (5) performs a fitting algorithm to determine a second polynomial based on the second sensor dataset to approximate the road marking (17) at the second measurement time. The second observation state vector includes the coefficients of the second polynomial. - Based on at least one motion parameter and a first observed state vector, calculate a predicted state vector for a second measurement time, the predicted state vector including coefficients of a polynomial used to approximate the road marking (17) at the second measurement time; and - Based on the predicted state vector and the second observed state vector, a corrected state vector for the second measurement time is generated, the corrected state vector including the coefficients of a polynomial used to approximate the road marking (17) at the second measurement time.

2. The method according to claim 1, Its features are, - The computing unit (5) is used to generate two or more first sampling state vectors for the first measurement time based on the first observation state vector and at least one parameter describing the multidimensional distribution; - The predicted state vector is calculated based on two or more first sampled state vectors.

3. The method according to claim 1 or 2, Its features are, The computing unit (5) is used for: - Calculate the prediction covariance matrix at the second measurement time based on the at least one motion parameter, the first observed state vector, and the predicted state vector; and - Generate a corrected covariance matrix for the second measurement time based on the predicted covariance matrix and the second observation state vector.

4. The method according to claim 1 or 2, Its features are, - By using the environmental sensor system (4), multiple sensor datasets depicting road markings (17) at corresponding continuous measurement times are generated, wherein the multiple sensor datasets include a first sensor dataset and a second sensor dataset; - For each measurement time, determine at least one motion parameter characterizing the motion of the environmental sensor system (4) relative to the road markings (17) between the corresponding measurement time and the corresponding subsequent measurement time; - For each measurement time, the computing unit (5) is used to generate the corresponding observation state vector of the geometric description road marking (17) based on the corresponding sensor dataset at the corresponding measurement time; - For each measurement time other than the final measurement time, the calculation unit (5) is used for: - Based on at least one corresponding motion parameter and the observed state vector at the corresponding measurement moment, calculate the predicted state vector for the corresponding subsequent measurement moment; and - Generate the correction state vector for the corresponding subsequent measurement time based on the predicted state vector and the observed state vector for the corresponding subsequent measurement time.

5. A method for at least partially automatically guiding a motor vehicle, the method comprising the steps of: - The method for road marking detection according to any one of claims 1 to 4, wherein the environmental sensor system (4) is mounted on the motor vehicle (1); and - The motor vehicle is guided at least partially automatically based on the corrected state vector (1).

6. The method according to claim 5, Its features are, - The computing unit (5) is used to generate electronic map data representing the environment of the motor vehicle (1) based on the corrected state vector; and - Automatically guide motor vehicles at least partially based on electronic map data (1).

7. The method according to any one of claims 5 or 6, Its features are, Based on the corrected state vector, one or more control signals for guiding the motor vehicle (1) are generated at least partially automatically.

8. A road marking detection system, comprising an environmental sensor system (4) configured to generate a first sensor dataset depicting a road marking (17) at a first measurement time, and to generate a second sensor dataset depicting the road marking (17) at a second measurement time, the environmental sensor system (4) comprising a lidar system; Its features are, - The road marking detection system (2) includes at least one motion sensor (6) configured to determine at least one motion parameter characterizing the motion of the environmental sensor system (4) relative to the road marking (17) between a first measurement time and a second measurement time; and - The road marking detection system (2) includes a computing unit (5) configured as follows: - Based on the first sensor dataset, a first observation state vector is generated to geometrically describe the road marking (17) at the first measurement time. The first sensor dataset includes a point cloud generated based on the sensor signals of the lidar system. The computing unit (5) performs a fitting algorithm to determine a first polynomial based on the first sensor dataset to approximate the road marking (17) at the first measurement time. The first observation state vector includes the coefficients of the first polynomial. - Based on the second sensor dataset, a second observation state vector is generated to geometrically describe the road marking (17) at the second measurement time. The second sensor dataset includes a point cloud generated based on the sensor signals of the lidar system. The computing unit (5) performs a fitting algorithm to determine a second polynomial based on the second sensor dataset to approximate the road marking (17) at the second measurement time. The second observation state vector includes the coefficients of the second polynomial. - Calculate a predicted state vector at a second measurement time based on at least one motion parameter and a first observation state vector, the predicted state vector including coefficients of a polynomial used to approximate the road markings at the second measurement time; as well as - Based on the predicted state vector and the second observed state vector, a corrected state vector for the second measurement time is generated, the corrected state vector including coefficients of a polynomial used to approximate the road markings at the second measurement time.

9. An electronic vehicle guidance system comprising a road marking detection system (2) according to claim 8 and a control unit, the control unit being configured to automatically generate at least partially, automatically, one or more control signals for guiding a motor vehicle (1) based on a correction state vector.

10. A motor vehicle comprising an electronic vehicle guidance system (3) according to claim 9 or a road marking detection system (2) according to claim 8.

11. A computer program product comprising instructions, wherein the instructions: - When performed by the road marking detection system (2) according to claim 8, the road marking detection system (2) performs the method according to any one of claims 1 to 4; or - When performed by the electronic vehicle guidance system (3) according to claim 9, the electronic vehicle guidance system (3) performs the method according to any one of claims 5 to 7.

12. A computer-readable storage medium for storing a computer program product according to claim 11.

Images