Method for automated parking of a vehicle

Abstract

Method for automated parking of a vehicle using a trajectory (T) previously learned at a learning time, characterized in that - when an automated parking function is activated at an activation time, the vehicle is located, - depending on the position (POS) of the vehicle, the learned trajectory (T) is converted into a coordinate system of a synthetic aperture radar image (SARB1, SARB2), - one synthetic aperture radar image (SARB1, SARB2) of a region in which the vehicle is located is received from a server for the learning time and the activation time, - a difference synthetic aperture radar image (DSARB) is determined based on differences (U1, U2) between the two synthetic aperture radar images (SARB1, SARB2), - an intersection between the learned trajectory (T) and the difference synthetic aperture radar image (DSARB) is determined and - the validity of the learned trajectory (T) is determined based on the intersection.

Description

[0001] The invention relates to a method for the automated parking of a vehicle using a previously learned trajectory.

[0002] From DE 10 2014 014 219 A1, a method for operating a driver assistance system for carrying out a parking maneuver of a vehicle is known. - a vehicle environment is recorded and stored during a learning drive, - a trajectory to a target position is determined and stored based on the recorded vehicle environment, - During the learning drive, obstacles on the determined trajectory are detected and stored and classified as passable and impassable obstacles for the vehicle, - the determined trajectory is adapted to the classified obstacles in such a way that impassable obstacles are avoided during the parking process.

[0003] Furthermore, DE 10 2019 214 628 A1 discloses a method which includes validating first object information and classification information from the detection of a vehicle's static environment using second object information and classification information from satellite-based environment detection. In this process, a fused image is generated from a satellite image in the visible wavelength range and a synthetic aperture radar image in order to derive the second object information and classification information.

[0004] Document DE 10 2017 107 626 A1 describes a method for detecting available parking spaces. Radar data from a parking environment is used as input for a neural network model. This network is trained to estimate parking space boundaries and approximate them as splines. A vehicle computer system uses these spline estimates to identify available parking spaces, with the estimates being continuously updated as the vehicle drives through the parking environment.

[0005] Document WO 2008 / 029038A1 discloses a device and a method for estimating the dimensions of a parking space. A synthetic aperture radar (SAR) is used to generate a complex image of the parking space. This complex SAR image is then processed to obtain a real-valued image, from which the length and depth values ​​of the parking space are extracted.

[0006] Document US 9959647B1 describes technologies for modeling activity patterns in remote sensing imagery using geospatial-temporal graphs (GSTs). Objects in the imagery are represented as nodes, with undirected edges representing spatial relationships (e.g., distance) and directed edges representing temporal changes. A distinction is made between persistent objects (e.g., buildings) and ephemeral objects (e.g., vehicles) that indicate activity, enabling the analysis of activity patterns over time.

[0007] Document DE 10 2011 109 492 A1 discloses a driving assistance device that records a traveled distance (trajectory) using wheel and steering angle sensors. The device uses a satellite navigation unit and digital map data to check whether the vehicle is in an area with limited navigation capabilities. If so, the recording of a trajectory is started automatically or at the driver's suggestion to enable the route to be driven later.

[0008] Document DE 10 2016 124 888 A1 discloses a method for assisting a driver with parking, based on a training and a playback mode. In training mode, image-based reference information is recorded using at least one camera during a parking trajectory controlled by the driver. In the subsequent playback mode, current camera images are compared with the stored reference information to determine the vehicle's position relative to the learned trajectory.

[0009] The invention is based on the objective of providing a novel method for the automated parking of a vehicle.

[0010] The problem is solved according to the invention by a method which has the features specified in claim 1.

[0011] Advantageous embodiments of the invention are the subject of the dependent claims.

[0012] In the automated parking process using a trajectory previously learned at a learning point, the vehicle is located when an automated parking function is activated at an activation point. Depending on the vehicle's position, the learned trajectory is converted into a coordinate system of a synthetic aperture radar image. Furthermore, a synthetic aperture radar image (SAR) of the region where the vehicle is located is received from a server for both the learning point and the activation point. A differential synthetic aperture radar image is then determined based on the differences between these two images.Furthermore, an intersection between the learned trajectory and the difference synthetic aperture radar image is determined, whereby the validity of the learned trajectory is determined based on the intersection.

[0013] Modern driver assistance systems include semi-autonomous parking functions that, after an initial learning of a route or trajectory, automatically follow it using data acquired by vehicle sensors. Due to the limited detection and resolution range of these sensors, the validity of the learned trajectory cannot be guaranteed at the time of learning. This can lead to significant corrections or even the termination of the parking function when the trajectory is subsequently followed. A trajectory can become invalid if, for example, structural changes, vegetation, parked vehicles, etc., alter the trajectory required for the parking maneuver.

[0014] The present method makes it particularly advantageous to precondition the automated parking function to determine whether a trajectory recalculation is necessary, thus enabling the execution of smooth, non-abrupt steering maneuvers during automated parking. The method also enables vehicle-independent three-dimensional object detection and mapping.

[0015] Since SAR is an atmospheric radar system which, through the artificial generation of a large aperture, enables a very high spatial resolution of, for example, 0.25 m, it offers the advantage over satellite-based camera systems that valid resolutions and object detections can be achieved even under overcast skies and bad weather conditions, so that the procedure can be carried out regardless of the weather and cloud cover in the atmosphere.

[0016] Furthermore, SAR allows the measurement of so-called acoustically hard objects, unlike camera-based systems, which, for example, consider vegetation as a whole, even though slight contact and / or driving over it is possible. Depending on the selected frequency band, leaves, etc., are not taken into account by SAR. This means that, depending on the selected frequency band, different materials, such as vegetation, stone, etc., can be considered.

[0017] Synthetic aperture radar images are also freely available and are updated cyclically, for example every 30 minutes. This enables reliable operation of the system.

[0018] Synthetic aperture radar images are also not locally bound but are generally available globally, so the procedure can be carried out reliably on a global scale.

[0019] Exemplary embodiments of the invention are explained in more detail below with reference to drawings.

[0020] This shows: Fig. 1. Schematically illustrates the process of an automated parking procedure for a vehicle. Fig. 2 schematically a visualization of a first procedural step of the procedure according to Fig. 1, Fig. 3. Schematic visualization of a third procedural step of the procedure according to Fig. 1, Fig. 4 schematically a visualization of a fourth process step of the procedure according to Fig. 1, Fig. 5 schematically a visualization of a fifth procedural step of the procedure according to Fig. 1, Fig. 6 schematically a visualization of a sixth procedural step of the procedure according to Fig. 1 and Fig. Figure 7 schematically shows a visualization of a seventh and an eighth process step of the procedure according to Fig. 1.

[0021] Corresponding parts are marked with the same reference symbols in all figures.

[0022] In Fig. Figure 1 shows a sequence of a possible embodiment of a method according to the invention for the automated parking of a vehicle using a trajectory T previously learned at a learning point, with several process steps S1 to S9. Fig. Figures 2 to 7 show visualizations of a first process step S1 ( Fig. 2) and a third to seventh process step S3 to S7 ( Fig. 3 to Fig. 7).

[0023] The procedure includes an approach for validating trajectories T initially learned at the time of learning. For example, in the context of automated parking functions, trajectories T can be initially learned by a driver of the vehicle and later automatically followed by the vehicle using data acquired by vehicle sensors.

[0024] For example, on private properties, a high variation in objects is to be expected, for instance due to structural changes. Therefore, the initially learned trajectory T may be faulty or invalid.

[0025] There is therefore a risk that the vehicle's sensors, for example due to limited sensor resolution, may only be able to detect such changes in the vehicle's surroundings to a limited extent, or not at all, in the immediate vicinity of the object in question, or due to obstructions. As a result, a smooth descent of trajectory T without abrupt steering and / or braking maneuvers would not be possible during the automated parking function, or the automated parking function would abort the driving task.

[0026] To verify the validity of the initially learned trajectory T, synthetic aperture radar images SARB1 and SARB2 are used. These are acquired using a synthetic aperture radar, also known as synthetic aperture radar (SAR).

[0027] In the present procedure, in a first process step S1, the trajectory T is initially taught at a learning time upon initiation by the driver of the vehicle.

[0028] Subsequently, in a second process step S2, the function for automated parking is activated at an activation time, i.e., for the automated and monitored driving along the trajectory T.

[0029] In a third process step S3, the vehicle is localized and, depending on the vehicle's position POS, the learned trajectory T is converted into a coordinate system of the synthetic aperture radar image SARB1, SARB2. The localization is performed, for example, using GPS data.

[0030] In a fourth process step S4, a synthetic aperture radar image SARB1 of a region in which the vehicle is located is retrieved or received from a server, for example a backend, for the learning time, and a synthetic aperture radar image SARB2 of the region in which the vehicle is located for the activation time.

[0031] In a fifth process step S5, a difference synthetic aperture radar image DSARB is determined based on differences U1, U2 between the two synthetic aperture radar images SARB1, SARB2.

[0032] In a sixth process step S6, an intersection between the learned trajectory T and the difference synthetic aperture radar image DSARB is determined.

[0033] Subsequently, in a seventh procedure step S7, the validity of the learned trajectory T is determined based on the intersection.

[0034] In a branch V, it is checked whether the validity is greater than a predefined threshold. If this is not the case, represented by a "no" branch N, an eighth process step S8 determines an alternative trajectory AT if the predefined validity threshold is not met. In a ninth process step S9, the determined alternative trajectory AT is then transmitted to the automated parking function and used for supervised parking or trajectory driving.

[0035] If, on the other hand, the validity is greater than the specified threshold, represented by a yes branch J, the initially determined trajectory T is used in the ninth procedure step S9 for supervised parking or trajectory driving.

Claims

[1] Method for automated parking of a vehicle using a trajectory (T) previously learned at a learning time, characterized by , that - when an automated parking function is activated at an activation time, the vehicle is located, - depending on the position (POS) of the vehicle, the learned trajectory (T) is converted into a coordinate system of a synthetic aperture radar image (SARB1, SARB2), - each a synthetic aperture radar image (SARB1, SARB2) of a region in which the vehicle is located, is received from a server for the learning time and the activation time, - a difference synthetic aperture radar image (DSARB) is determined based on differences (U1, U2) between the two synthetic aperture radar images (SARB1, SARB2), - an intersection between the learned trajectory (T) and the difference synthetic aperture radar image (DSARB) is determined and - the validity of the learned trajectory (T) is determined based on the intersection. [2] Method according to claim 1, characterized by , that if a predefined validity threshold is not met, an alternative trajectory (AT) is determined and transmitted to the automated parking function. [3] Method according to claim 1 or 2, characterized by , that the activation of a function for learning a trajectory (T) is manually activated.

Images