Methods for estimating the geometry of a motion path

Abstract

Method for estimating the geometry of a movement path of an ego vehicle (101), in which sensor data encompassing detection points are acquired using a sensor and the geometry of the movement path is estimated based on the sensor data, wherein targets along the movement path are determined based on the sensor data, location points for the respective position of the targets and the sensor are defined, several polygons, in particular continuously, are determined based on the location points, so that a polygon mesh is created which is used to estimate the geometry of the movement path, characterized in that the method comprises the following process steps: Step 1: an ego point (S) is assigned to the acquired detection points, which indicates the sensor position, Step 2: the polygon mesh is created based on the detection points, Step 3: the polygons are marked, Step 4: the edges of the polygons areStep 6: The edges are removed according to their weighting, so that in particular the edges where the movement path is not possible or only possible to a limited extent are removed. Step 7: Possible circumnavigations (UR) are detected by recognizing cycles within the polygon mesh. Step 8: The movement path through the polygon mesh is determined based on the detected cycles.

Description

[0001] The present invention relates to a method for estimating the geometry of a movement path of a means of locomotion, in particular an ego-vehicle, and to a system for carrying out the method according to the invention. Technological background

[0002] Modern means of transportation, such as cars or motorcycles, are increasingly equipped with driver assistance systems. These systems use sensors to perceive the environment, recognize traffic situations, and support the driver, for example, by intervening with braking or steering, or by issuing visual or audible warnings. Radar sensors, lidar sensors, camera sensors, and similar devices are regularly used as sensor systems for environmental perception. The sensor data collected by these sensors allows conclusions to be drawn about the environment, enabling, for example, object and / or environment classification. Furthermore, environmental perception is virtually indispensable in the field of (partially) autonomous driving. By processing the sensor and environmental data, assistance functions can then be executed, which, for example,Accidents with other road users can be avoided or complicated driving maneuvers made easier by supporting or even completely taking over the driving task or vehicle control (partially / fully automated).

[0003] The detection and estimation of road edges or boundaries is a fundamental function in modern driver assistance systems and is used, for example, in adaptive cruise control (ACC) or emergency braking systems (EBA) to estimate the trajectory or path of the respective vehicle. The road boundary is typically expressed as a clothoid. In modern road construction, a clothoid is defined, for example, as the transition curve between a straight line and a bend. The specific shape of the clothoid can be estimated, for example, based on static targets or objects located near the road boundary and detected by radar measurements. Such an estimation of the road boundary is also performed using a Kalman filter, which, for example,This serves to reduce errors in real-world measurements and to provide estimates for non-measurable system variables.

[0004] Radar sensors are frequently used in modern vehicles to perceive their surroundings. This technology relies on the emission of focused electromagnetic waves (radar waves or radar signals) and their reflection, for example, by other road users, obstacles on the road, or roadside structures. Distinguishing between obstacles or objects, such as other road users, traffic signs, road markings, roadside structures, guardrails, and the like, is crucial for determining when, for example, braking or steering interventions should be initiated. Printed state of the art

[0005] US Patent 6,751,547 B2 discloses a method for accurately estimating the geometry of a vehicle's forward path based on a two-clothoid road model. This method involves collecting road data provided by a camera or radar system. A measurement transfer function of the two-clothoid road model is then calculated from this data. The clothoid coefficients for the near and far ranges are estimated simultaneously, and the vehicle's forward path is calculated based on the data provided by the two-clothoid model.

[0006] US Patent 6,718,259 B1 discloses a method for estimating the geometry of a vehicle's forward path using an adaptive Kalman filter bank and a two-clothoid road model. This method employs multiple Kalman filters, with weighting factors provided for each filter that indicate the probability of the anticipated road geometry matching the road model assumed in the filter. After each filter has been assigned a weighted value, the weighted road models are fused using a fusion element to output a single weighted road model.

[0007] Furthermore, DE 10 2016 107 705 A1 discloses a method in which a path for an autonomous vehicle is planned using Delaunay triangulation, collecting all detections, including those of static and moving objects. The triangulation method is then used to search for triangular edges (triangular segments) lying on the road surface that meet a series of criteria, one of which is that each edge is not an edge at the road boundary. After finding the possible edges or paths, the vertices of these edges are extracted, and the path is planned based on these vertices. A function of the various path costs is then estimated, and the path with the lowest cost is selected.

[0008] Further methods for estimating the geometry of a movement path of a means of locomotion are known from DE 10 2015 013 084 A1 and DE 10 2013 201 796 A1.

[0009] Furthermore, DE 10 2018 215 448 B3 discloses a method for estimating the geometry of a movement path of a means of locomotion according to the preamble of claim 1.

[0010] One problem with known methods is that such algorithms often fail in urban scenarios with multiple road corridors. In such situations, the vehicle can typically choose from various road corridors or paths, for example, at an intersection, and each of these paths must be evaluated. However, the prior art algorithms do not provide additional information about whether the upcoming paths or roads are practical (i.e., easily navigable), maneuverable (i.e., the number of maneuvers required to complete the path), and passable (i.e., the driving corridor is wide enough for the vehicle to pass through). Since urban scenarios thus involve more than one path and each path must be evaluated, there is a particular interest in providing a method that overcomes these limitations. Object of the present invention

[0011] Starting from the prior art, the object of the present invention is now to provide a method in which the estimation of the geometry of the movement path, particularly in urban scenarios, is improved in a simple and cost-effective manner, thereby overcoming the disadvantages of the prior art. Solution to the task

[0012] The aforementioned problem is solved by the entire teaching of claim 1 and the dependent claim. Advantageous embodiments of the invention are claimed in the dependent claims.

[0013] In the inventive method for estimating the geometry of the path of a means of transportation, in particular a vehicle, sensor data is collected by means of a sensor, and the geometry of the path is then estimated based on this sensor data. This is done by identifying, in particular, static targets or objects along the path of motion using the sensor data and defining location points for the respective positions of the targets and the sensor. Furthermore, a polygon is determined based on these location points, for example, by using a definable number of location points as vertices of the polygon. In particular, the polygon can include location points on each side of the road. Subsequently, the polygon is used to estimate the geometry of the path of motion; that is, the inner road boundary points can be determined, for example, for both road boundaries. This determination can be implemented, for example, in the form of an algorithm.According to the invention, the inner boundary of the road can be determined or calculated based on sensor measurements. Furthermore, by determining multiple polygons, particularly continuously, so that an entire polygon network is generated, this network can be used to estimate the geometry of the movement path. The method according to the invention comprises the following steps: Step 1, in which an ego point is assigned to the detected points, indicating the sensor position; Step 2, in which the polygon network is created based on the detected points; Step 3, in which the polygons are marked, e.g.,counted and numbered or the like; step 4, in which the edges of the polygons are weighted, in particular depending on at least one edge parameter; step 6, in which the edges are removed according to their weighting, in particular removing those edges where the path of movement is not possible or only possible to a limited extent, for example, by falling below or exceeding a certain weighting value or weighting factor; step 7, in which possible circuits are detected by recognizing cycles within the polygon network; and step 8, in which the path of movement through the polygon network is finally determined based on the detected cycles. The method or algorithm according to the invention offers the advantage that the search for road corridors, circular arcs, and cross streets in the static environment is particularly facilitated.Furthermore, the quality of each road corridor is estimated, and impassable corridors can be eliminated, thus improving selection accuracy. Such features were not present in known methods or algorithms for road corridor detection. In addition, the algorithm exhibits low computational complexity and is easy to implement or retrofit. Moreover, the method is independent of side classification or the calculation of previous radar cycles. Therefore, the method according to the invention can also be used for the classification or estimation of trajectory or movement path geometry in dynamic scenarios, such as road widening or narrowing, road construction work, missing guardrails, or the like.

[0014] According to a preferred embodiment, a radar sensor is provided as the sensor, which generates radar detections resulting from the emission of focused electromagnetic waves (radar waves or radar signals) and their reflection, e.g., from other road users, obstacles on the roadway, or roadside structures. However, other sensors known from the prior art can also be provided, such as camera, ultrasonic, or lidar sensors, whose detection points are fed into the algorithm.

[0015] It has proven particularly advantageous if the polygon is a triangle, as this can be calculated very easily with minimal computational effort. Similarly, an entire polygon network can be created using triangle formations. This can be done simply using Delaunay triangulation or Delaunay triangle formation, as already described in DE 10 2018 215 448 B3.

[0016] Furthermore, the procedure can include step 5, in which the edges are mapped and labeled as terminals. This allows the procedure to be made even more secure and simplified.

[0017] Preferably, the edge parameter can be the edge length, the edge angle, the angle of the road surface and / or the centroid angle.

[0018] The movement path can be conveniently determined by selecting the shortest path through the polygon mesh. This simplifies the selection process accordingly.

[0019] According to a specific implementation of the method, the detection points can be grouped and represented as polylines, allowing the creation of a restricted polygon mesh based on these polylines, which enclose the grouped detection points. By using a restricted polygon mesh, or restricted Delauney triangulation, the method can be significantly simplified, thereby improving computational reliability and saving computing resources.

[0020] Furthermore, the present invention comprises a system for estimating the geometry of the movement path of a means of locomotion. For this purpose, the system includes a sensor for generating sensor data, in particular a radar sensor for generating radar detection points, and a control unit. Based on the sensor data or the detection points, the geometry of the movement path is then estimated by the control unit being configured to receive the sensor data and process it further according to the method sequence of the invention. Advantageously, the control unit can be configured to access actuators of the ego vehicle in order to perform corresponding driving and assistance functions in a supporting and / or (semi-)autonomous manner.

[0021] Advantageously, the method according to the invention can be a purely computer-implemented method, wherein the term "computer-implemented method" within the meaning of the invention describes a process plan or procedure that is implemented or carried out using a computer. The computer, such as a computer, a computer network, a control unit (e.g., ECU or Electronic Control Unit or ADCU or Assisted & Automated Driving Control Unit), a computing device therein, or another programmable device known from the prior art, can process the relevant data by means of programmable calculation instructions. With regard to the method or system according to the invention, the essential properties can be achieved, for example, by a new program, new programs, an algorithm, or the like, which is executed on the control unit. List of characters

[0022] The invention will now be explained in more detail using practical embodiments. The figures show: Fig. 1 a simplified schematic representation of a vehicle in which an estimation of the geometry of a movement path is carried out using the method according to the invention; Fig. 2 a simplified schematic representation of static radar detections in a point cloud; Fig. 3 a simplified schematic representation of the point cloud from Fig. 2 with marked Ego-Point S (Step 1); Fig. 4 a simplified schematic representation of the point cloud from Fig. 3 with successful Delaunay triangulation over all static points (step 2); Fig. 5 a simplified schematic representation of the point cloud from Fig. 4 with numbered triangles (step 3); Fig. 6 a simplified schematic representation of the point cloud from Fig. 4 with numbered triangles and weighted edges marked as letters (step 4); Fig. 7 a simplified schematic representation of the point cloud from Fig. 5 as an indirect and weighted graph, where the end edges of the triangulation are labeled as terminals t1-t6 (step 5); Fig. 8 a simplified schematic representation of the point cloud from Fig. 6 and of the diagram from Fig. 7, removing impractical edges of the triangulation in the graph (step 6); Fig. 9 a simplified schematic representation of the recognition of circles by identifying cycles in the graph (step 7); Fig. 10 a simplified schematic representation of the final design of the shortest path from the ego point S to t3 Fig. 19 (Step 8); Fig. 11 a simplified schematic representation of a design of a restricted Delaunay triangulation; Fig. 12 a simplified schematic representation of a design of the restricted Delaunay triangulation from Fig. 11 as an indirect graph (weightings are not shown for clarity); Fig. 13 a simplified schematic representation of the restricted Delaunay triangulation from Fig. 12, where the shortest path is represented by triangulation through the removal of paths, as well as Fig. 14 a simplified schematic representation of a design of an intersection detection system.

[0023] Reference number 101 in Fig. 1 denotes an ego vehicle with a control unit 102 (ECU, Electronic Control Unit or ADCU, Assisted and Automated Driving Control Unit) that enables sensor control, sensor data fusion, environment and / or object recognition, trajectory planning, and / or vehicle control, particularly (semi-)autonomously. For vehicle control, the control unit 102 can access various actuators (steering 103, motor 104, brake 105). Furthermore, the ego vehicle 1 has various sensors for environment perception (Lidar 106, camera 107, radar 108, and ultrasonic sensors 109a-109d). Advantageously, the sensor data can be used for environment and object recognition, enabling the implementation of various assistance functions, such as emergency brake assist (EBA), adaptive cruise control (ACC), lane keeping assist (LKA), or the like.Furthermore, the execution of the assistance functions can also be carried out via the control unit 102 or another control unit provided for this purpose. The trajectory planning is based on an estimate of the geometry of the movement path according to the method according to the invention.

[0024] In the method according to the invention, the geometry of the path of motion is estimated by determining or calculating the inner road boundary based on sensor measurements. Exemplary embodiments are described below in which a radar sensor is used. However, other sensors known from the prior art, such as lidar or camera sensors, are also expressly included. In each scan cycle, the radar sensor provides a list of objects or targets with the corresponding relative position of each target and their velocity relative to the ego-vehicle, which includes the radar sensor and corresponds to location point 0 (x = 0; y = 0). Based on the relative velocity to the ego-vehicle, the objects are then classified as static or moving targets. For estimating the road boundary, however, only the static targets, each of which can be assigned a location point, are used in this case.The estimation is then performed by calculating or inserting a polygon mesh or a Delaunay triangulation (Delaunay triangle formation or Delaunay triangulation) based on the position of the static targets or their location points and that of the ego vehicle. The creation of the polygon mesh or the Delaunay triangulation is carried out in particular by a method or method design according to DE 10 2018 215 448 B3, to which direct reference is therefore made.

[0025] The process according to the invention is described below using the following as an example. Fig. 2-10 shown. Fig. Figure 2 initially displays a point cloud of radar detections. The aim is to identify all drivable paths, assessing the feasibility of each path and, if possible, creating a scene map. The first step involves adding an ego point (here denoted as S) representing the sensor position ( Fig. 3). In the second step 2, the polygon mesh creation or Delaunay triangulation then takes place ( Fig. 4), which is placed over all static points. Then the polygons or triangles are marked (step 3), e.g., numbered or counted, as in Fig. Figure 5 illustrates this. In the next step (step 4), the edges of the triangles can then be weighted by considering, for example, edge parameters of the polygons such as edge length, road edge angle, road surface angle, centroid angle, and the like. An example is shown in Fig. Figure 6 shows an example of Delaunay triangle triangulation with numbered triangles and weighted edges. The edges of the triangulation can then be mapped and labeled as terminals t1-t6, according to step 5. Such a network of triangles is represented as an indirect and weighted graph in Fig. 7 shown.

[0026] In practical terms, edges can then be removed where movement is not possible or only possible with disadvantages or conditions (at high cost, with a borderline weighting factor, or the like) (step 6), e.g., if the edge is too narrow or it is not possible to perform certain driving maneuvers (e.g., too steep a turn or curve maneuver with regard to the ego speed). Fig. Figure 8 illustrates the removal of impractical edges in the triangulation (left) and in the graphic (right). According to Fig. For example, edges b and e (marked by oval markings) are removed (and thus also the path through triangle 3), as these would require excessively steep driving maneuvers.

[0027] In step 7, the cycles in the graph can then be specified, in which each cycle represents a possible circumnavigation in the static environment, as in Fig. Figure 9 shows (circling the UR by detecting cycles in the graph). Furthermore, according to step 8, various heuristics can be used to perform the search within the graph or the tree structure, e.g., breadth-first search, greedy search, and the like. Statistics can be collected during the traverse of the tree (maneuverability). Each path in the tree is a possible route or road corridor (candidate), where the depth of a subtree reflects the number of supported data points at an intersection. Fig. Figure 10 shows the mapping of the shortest path from S to t3 to the triangulation. Based on the cycles, the shortest path through the polygon network or the triangulation can then be calculated, allowing this information to be used for trajectory planning. The algorithm significantly simplifies the search for suitable road corridors, circular arcs, and cross streets in the static environment and also provides an estimate of the quality of the respective movement path or road corridor, whereby impassable corridors can be eliminated. The final design of the shortest path from S to t3 is shown in Fig. 10 shown.

[0028] Furthermore, the present algorithm can be extended, particularly for urban street scenarios. The detected objects or detection points are grouped (clustered) and represented as polylines. Subsequently, a restricted Delaunay triangulation can be applied, in which the cluster polylines are considered bounded edges in the triangulation process. A restricted Delaunay triangulation is a generalization of Delaunay triangulation that forces certain required segments into the triangulation, with the polylines enclosing the grouped or clustered objects. Such a clustering process can, for example, be used to describe the... Fig. Output the restricted Delaunay triangulation shown in Figure 11, where certain areas have been restricted (ES restriction). The polylines of the object clusters are treated as restricted edges in the triangulation process.

[0029] The described procedure steps 1-5 can then be applied to restricted Delaunay triangulation. The result of the application to restricted Delaunay triangulation is shown as an indirect graph in Fig. Figure 12 is shown (without weighting for clarity). By subsequently removing the restricted edges belonging to the graph objects (step 6), the weighted shortest path search algorithms from ego point S to terminals t1, ..., t7 can be applied, as shown in Fig. Figure 13 shows the edges to be removed, marked by scissors. The path can then be evaluated as it traverses the graph, for example, by accumulating weights (e.g., centroid lengths). For example, as in Fig.As shown in Figure 14, the paths from S to t5 and t2 are analyzed, with triangle 22 being a crossing point KP, since two different paths pass through the two edges of triangle 22 (see transition from 21 to 23 and from 21 to 20). Reference symbol list 101 Ego Vehicle 102 Control unit 103 Steering 104 engine 105 brake 106 Lidar sensor 107 Camera 108 radar sensor 109a-109d Ultrasonic sensor S Ego Point t1-t7 Terminal KP Intersection Point ES restriction QUOTES INCLUDED IN THE DESCRIPTION

[0000] This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited patent literature

[0000] US 6751547 B2

[0005] US 6718259 B1

[0006] DE 102016107705 A1

[0007] DE 102015013084 A1

[0008] DE 102013201796 A1

[0008] DE 102018215448 B3 [0009, 0015, 0024]

Claims

[1] Method for estimating the geometry of a movement path of an ego vehicle (101), in which sensor data comprising detection points are recorded by means of a sensor and the geometry of the movement path is estimated on the basis of the sensor data, wherein targets along the movement path are determined on the basis of the sensor data, location points for the respective position of the targets and the sensor are determined, several polygons are determined on the basis of the location points, in particular continuously, so that a polygon network is created which is used to estimate the geometry of the movement path, characterized by that the procedure comprises the following procedural steps: Step 1: the recorded detection points are assigned an ego point (S), which indicates the sensor position, Step 2: the polygon network is created based on the detection points, Step 3: the polygons are marked, Step 4: the edges of the polygons are weighted, in particular depending on at least one edge parameter, Step 6: the edges are removed according to their weighting, so that in particular the edges are removed where the movement path is not or only partially possible, Step 7: Detecting possible orbits (UR) by detecting cycles within the polygon mesh, and Step 8: Determine the movement path through the polygon mesh based on the detected cycles. [2] Method according to claim 1, characterized by that at least one lidar sensor (106), a camera (107), a radar sensor (108) and / or an ultrasonic sensor (109a-109d) is provided as the sensor. [3] Method according to claim 1 or 2, characterized by that the polygon is a triangle or the polygons are triangles. [4] Method according to claim 3, characterized bythat the polygon mesh is determined using a Delaunay triangulation. [5] Method according to at least one of the preceding claims, characterized by that the method further comprises step 5, after which the edges are marked as terminals t1- t6. [6] Method according to at least one of the preceding claims, characterized by that the edge length, the edge angle, the angle of the road surface and / or the centroid angle are provided as edge parameters. [7] Method according to at least one of the preceding claims, characterized by that the movement path is determined by selecting the shortest path through the polygon mesh. [8] Method according to at least one of the preceding claims, characterized bythat detection points are grouped and represented as polylines, so that a constrained polygon mesh is created from the polylines, with the polylines enclosing the grouped detection points. [9] Method according to at least one of the preceding claims, characterized by , that [10] System for estimating the geometry of a movement path of an ego vehicle (101), with sensor data comprising at least one sensor for generating detection points, and a control device (102), wherein the estimation of the geometry of the movement path is carried out by the control device (102) on the basis of the sensor data of the sensor, in that the control device (102) carries out the estimation of the geometry of the movement path by means of a method according to one of the preceding claims.

Images