METHOD AND DEVICE FOR DETERMINING FALSE-POSITIVE DETECTIONS OF A LIDA SENSOR

Description

[0001] The invention relates to a method for determining false-positive detections of a lidar sensor during a scanning process of a vehicle environment.

[0002] The invention further relates to a device for determining false-positive detections of a lidar sensor during a scanning process of a vehicle environment.

[0003] From DE 199 47 593 A1, a radar device for a vehicle, which is to be attached to a vehicle, is known. The radar device has a radar beam scanning device for scanning a scanning area with a radar beam parallel to a road surface on which the vehicle is traveling. Furthermore, the radar device has an object detection device for receiving reflected waves of the radar beam in order to create a detected object image based on the reflected waves, wherein the object detection device includes a ghost echo detection device for determining whether a generated detected object image is a ghost echo.

[0004] From EP 3 121 620 A1, a method for segmenting a 3D point cloud of measurements, recorded with a laser sensor in the presence of an aerosol cloud generated during helicopter landing, is known. The method involves clustering the 3D point cloud based on a local point density, determining temporal changes in the position, orientation, and shape of the clusters, and identifying the association of a cluster with the aerosol cloud based on these changes.

[0005] From the publication by Koch, R., May, S., and Nüchter, A.: "Effective distinction of transparent and specular reflective objects in point clouds of a multi-echo laser scanner", 18th International Conference on Advanced Robotics (ICAR), 2017, pp. 566-571, ISBN 978-1-5386-3158-4, a method for detecting transparent and specular reflective objects in a point cloud from a laser scanner is known. In this method, discontinuities in the distance values ​​of the points in the point cloud are identified by evaluating their distances. Upon finding such discontinuities, the corresponding points are used to locate flat, square surfaces. The identified flat surfaces and the intensities of the reflections are then used to distinguish between transparent and specular surfaces.

[0006] The invention is based on the objective of providing a novel method and a novel device for determining false-positive detections of a lidar sensor.

[0007] The problem is solved according to the invention by a method which has the features specified in claim 1 and by a device which has the features specified in claim 6.

[0008] Advantageous embodiments of the invention are the subject of the dependent claims.

[0009] In a method for evaluating reflections of lidar pulses from a lidar sensor, in which the lidar pulses are sent by the lidar sensor in different directions of a scan area of ​​the lidar sensor during a scanning process of an environment, and reflections of the lidar pulses are detected by the lidar sensor, and in which clusters are generated by clustering the reflections, it is provided according to the invention that that the environment is a vehicle environment, that for each laser pulse which is reflected back to the lidar sensor in the scan area, it is checked whether it is reflected back multiple times at different distances, that for the laser pulses reflected back multiple times, the reflections received first are determined as the first reflections, that the clustering is carried out based on the first reflections, and that if a distance evaluation of reflections from a generated cluster shows that the laser pulses are reflected there at an at least approximately homogeneous reflective surface, further reflections following the respective first reflection are marked as false positive detections.

[0010] This method reliably identifies false positive detections, also known as ghost targets or ghost echoes, by a lidar system during a vehicle scan. This makes it possible to avoid serious limitations of lidar systems in real-world traffic situations caused by reflective materials such as the windows of other vehicles. Consequently, the reliability of driver assistance systems, particularly those in automated, highly automated, or autonomous vehicles, can be increased.

[0011] In another possible embodiment of the method, a principal component analysis is performed to determine the surface of the clusters, distinguishing between clusters with planar (i.e., flat) surfaces, clusters with curved surfaces, clusters extending in exactly one spatial direction, and clusters extending in three spatial directions. Based on this distinction, the clusters can be classified in such a way that it can be easily determined, simply by virtue of this classification, whether a false positive detection is possible within a given cluster.

[0012] In another possible embodiment of the method, only clusters with a planar surface and clusters with a slightly curved surface, whose curvature corresponds to the curvature of a vehicle window pane, are characterized as approximately homogeneously formed reflective surfaces.

[0013] In another possible embodiment of the procedure, clusters that extend in exactly one spatial direction are characterized as clusters representing the edges of objects.

[0014] In another possible embodiment of the method, clusters extending in three spatial directions are characterized as clusters representing dust and / or fog and / or fine-grained structures.

[0015] An inventive device for evaluating reflections of lidar pulses comprises a lidar sensor and an evaluation unit, wherein the lidar sensor is configured to send the laser pulses in different directions of a scan area of ​​the lidar sensor during a scanning process of a vehicle environment and to detect reflections of the lidar pulses, and wherein the evaluation unit is configured The system aims to generate clusters by clustering the reflections, to check for each laser pulse that is reflected back to the lidar sensor within its scan area whether it is reflected back multiple times at different intervals, to determine the first reflections received for each of the multiple reflected laser pulses as the first reflections, to perform the clustering based on the first reflections, and then, if a distance evaluation of reflections from the generated cluster shows that the laser pulses are reflected at an at least approximately homogeneous reflective surface, to mark subsequent reflections following the respective first reflection as false positives.

[0016] The device reliably identifies false-positive detections by a lidar sensor during a vehicle's surroundings scan. This prevents serious limitations of lidar sensors in real-world traffic situations caused by reflective materials, such as the windows of other vehicles. Consequently, the reliability of driver assistance systems, particularly those in automated, highly automated, or autonomous vehicles, can be increased.

[0017] Exemplary embodiments of the invention are explained in more detail below with reference to drawings.

[0018] This shows: Fig. 1 schematically shows a lidar sensor, an object, and the path of laser radiation with diffuse reflection from the object; Fig. 2 schematically shows a lidar sensor, two objects, and the path of laser radiation with mirror-like reflection from one of the objects and diffuse reflection from the other object; Fig. 3 schematically shows a lidar sensor, two objects, a ghost object, and the path of laser radiation; Fig. 4 schematically shows a traffic scenario; Fig. 5 schematically shows a lidar sensor, two objects, and the path of reflected laser pulses; Fig. 6 schematically shows a lidar sensor, an object, and the path of reflected laser pulses with dust or fog between the lidar sensor and the object; Fig. 7 schematically shows a lidar sensor, a partially transmissive object, two non-transmissive objects, and the path of reflected laser pulses.Figure 8 schematically shows a lidar sensor, a partially transmissive object, three non-transmissive objects, and a path of reflected laser pulses. Figure 9 schematically shows a lidar sensor, two objects, and a path of reflected laser pulses. Figure 10 schematically shows a lidar sensor, two objects, another object formed by dust and / or fog, and a path of reflected laser pulses. Figure 11 schematically shows a lidar sensor, a partially transmissive object, two non-transmissive objects, and a path of reflected laser pulses.

[0019] Corresponding parts are marked with the same reference symbols in all figures.

[0020] In Figure 1 A lidar sensor 1, an object 2 and a path of laser radiation L 1 , L 2 during diffuse reflection at the object 2 are shown.

[0021] Lidar sensors 1 are for in Figure 4The vehicles 5 and their driver assistance systems, as described in more detail, for example for automated, especially highly automated or autonomous vehicles 5, are very important for capturing a vehicle's surroundings, since lidar sensors 1 enable a precise three-dimensional understanding of the traffic scene surrounding the vehicle 5. However, in order to be used in such a safety-critical scenario, it is crucial that Figure 11 The aim is to determine the false positive detections or false positive distance measurements of the lidar sensors. Failure to detect such false positives can lead to dangerous automatic interventions in the longitudinal and / or lateral control of the vehicle 5, which can result in dangerous situations, such as unjustified emergency braking.

[0022] Lidar sensors 1 are generally characterized by very low false-positive rates, i.e., low noise levels. With most target materials, the incident laser radiation L1, L2 is diffusely reflected, i.e., in all possible directions. This means that some of the light is reflected directly back into the lidar sensor 1, enabling precise distance measurement.

[0023] However, specular reflection can dominate on highly reflective surfaces. This is in Figure 2 further explained, whereby Figure 2 shows a lidar sensor 1, two objects 2, 3 and a path of laser radiation L 1 , L 2 with a mirror-like reflection at one of the objects 3 and a diffuse reflection at the other object 2.

[0024] The laser beam L1 is deflected away from the lidar sensor 1, where it may strike other non-reflective surfaces, such as object 2, as laser beam L2. The lidar sensor 1 does not measure the distance of the laser beam L1 to the surface of object 3, but rather the total length of a zigzag path formed by both laser beams L1 and L2.

[0025] The lidar sensor 1 receives no information indicating that the laser beam L 2 has been redirected, so it assumes a distance measurement along a straight line. This results in a false positive detection along an initial outgoing direction along the laser beam L 1, leading to the detection of a Figure 3 The ghost object shown, number 4, results.

[0026] This presents a fundamental problem which, if left unaddressed, can lead to critical errors. In real-world traffic scenarios, highly reflective surfaces frequently appear in the form of the windows of other vehicles (shown in...). Figure 4 Even if these objects do not reflect perfectly specularly, they still deflect at least some of the incident laser radiation L1. If the deflected laser radiation L2 then encounters a highly reflective secondary object 2, for example a traffic sign 8, a resulting indirect path can dominate a direct path, such as one formed by diffuse reflection at the window of a vehicle.

[0027] Such a situation is exemplified in Figure 4 depicted. Figure 4This shows a traffic scenario with a vehicle 5, which has the lidar sensor 1, an object 3 in front of it designed as another vehicle 6, another object 2 designed as a traffic sign 8 and a ghost object 4.

[0028] Such a ghost object 4 appears on a roadway FB in front of vehicle 5. Such false-positive detections can be consistent over extended periods, but can also exhibit unusual dynamics, as the precise position depends on the relative distance between the two vehicles 5, 6 and the distance between a primary and a secondary target, for example, the window pane 7 and the traffic sign 8. If such reflection effects are not detected, there is a risk that a driver assistance system or an automated driving system of vehicle 5 will react to the ghost object 4 and initiate an incorrect intervention in the longitudinal and / or lateral control of vehicle 5, for example, an emergency braking or an evasive maneuver.

[0029] To enable the detection of such reflection effects and, consequently, the reliable detection of ghost objects, a determination of false-positive detections by a lidar sensor 1 during a scan of a vehicle's surroundings is provided, which is based on the following Figures 5 to 11 is described.

[0030] This shows Figure 5 A lidar sensor 1 and two objects 2, 3 are involved when the object 3 located between object 2 and lidar sensor 1 is only partially struck by the laser radiation L 1 emitted by lidar sensor 1. This results in two reflected laser pulses R 1 , R 2 .

[0031] In Figure 6 The figure shows a lidar sensor 1, an object 2 and another object 3 formed by dust and / or fog between the lidar sensor 1 and the object 2.

[0032] Due to the dust and / or fog, two reflected laser pulses R 1 , R 2 are also generated.

[0033] Figure 7shows a lidar sensor 1, a partially transmissive object 3, for example a window pane 7 of a vehicle 6, two non-transmissive objects 2, 9.

[0034] Partially transmissive objects 3, such as glass windows, can lead to three different distance measurements. For example, direct diffuse reflection from defective and possibly dirty window panes 7 can occur, as illustrated by the reflected laser pulse R 1.

[0035] Specular reflection can also occur on the glass surface, leading to ghost reflections, represented by the reflected laser pulses R2 and R2'. In particular, the reflected laser pulse R2 is reported as a false-positive laser pulse R2' along the original beam direction.

[0036] Furthermore, transmission through the glass and diffuse reflection at the object 2, which is designed as a background object, can lead to a reflected laser pulse R 3.

[0037] In one embodiment not shown in detail, the length of the reflected laser pulse R2 can also be greater than the length of the reflected laser pulse R3. However, the direct diffuse reflection, represented by the reflected laser pulse R1, is always the first received reflection with the shortest distance.

[0038] In exemplary embodiments not shown in detail, other beam paths are also possible, for example a reflection from a back side of the window pane 7 after an impact on the object 2 designed as a background object, which, however, are of a higher order and significantly weaker.

[0039] To detect false positives, such as the reflected laser pulse R 2 ', it is necessary to first distinguish between the ones in the Figures 5 and 6 to distinguish between the scenarios presented. Such a distinction is made more difficult by the fact that in the Figure 7 In the depicted scenario, not every laser beam leads to a weak direct reflection according to the reflected laser pulse R 1 at the window pane 7.

[0040] This problem shows Figure 8 This will be explained in more detail using a more complex scene. It shows Figure 8 a lidar sensor 1, a partially transmissive object 3, three non-transmissive objects 2, 9, 10 and a course of reflected laser pulses R 1 to R 4 .

[0041] A direct diffuse reflection at the transmissive object 3, for example the window pane 7 of the vehicle 6, leads to a small number of first reflections on a surface of the object 3, where the first reflections are represented by squares.

[0042] For some laser beams, specular reflection at object 3 dominates, while for others, the transparent path at object 3 is dominant, resulting in reflections at objects 2 and 9. In some cases, the laser beams pass through object 3 in such a way that the first reflections only occur at objects 2 and 9.

[0043] Furthermore, second reflections of a laser beam are represented as triangles and third reflections as circles.

[0044] To determine false positive detections in the measurements, it is planned that the complete scan process carried out using the lidar sensor 1 will be searched for measurements with second and possibly higher reflections.

[0045] Subsequently, a simple Euclidean clustering algorithm is applied to all first reflections which occur in the Figures 9 to 11 represented as squares, applied to bring together nearby points.

[0046] For each extended cluster C1, C2 formed in this way, a principal component analysis is performed, resulting in three sorted eigenvalues ​​λ1 ≥ λ2 ≥ λ3. Clusters C1, C2 are in the Figures 9 to 11 A more detailed explanation.

[0047] Planar clusters C1, C2 with λ1, λ2 » λ λ3 and clusters C1, C2 with slightly curved surfaces are marked as candidates for reflective surfaces.

[0048] Clusters C1, C2, which extend in only one direction (λ1 » λ2, λ3), most likely result from object edges and are therefore not marked. Clusters C1, C2, which extend in all directions (λ1 ≈ λ2 ≈ λ3), most likely result from fog and / or dust and / or very fine-grained structures, such as trees or vegetation, and are also not marked.

[0049] False positive reflections or detections E are marked as measurements which meet the following two conditions: 1. The reflection results from a measurement along a laser beam passing through a volume spanned by a labeled cluster C1, C2 (shown in the Figures 9 to 11(by dashed outlines of the first reflections). 2. A distance to the reflection is greater than a distance to the marked cluster C1; that is, the reflection is located behind cluster C1, C2 from the perspective of the lidar sensor.

[0050] Figure 9 Figure 1 shows a lidar sensor 1, two objects 2 and 3, and a path of reflected laser pulses R1 and R2. A laser beam is partially reflected at one edge of object 2, with the remaining portion reflected at the other object. The reflection at the edge of object 2 is a first reflection, and the reflection at object 3 is a second reflection. Since there is no extended region with second reflections, only a cluster C1 in the area of ​​the first reflection is marked. The remaining reflections (represented by crosses), which are not followed by any further reflections, are not marked.

[0051] In Figure 10The diagram shows a lidar sensor 1, two objects 2 and 9, and a further object 3 formed by dust and / or fog between the lidar sensor 1 and the objects 2 and 9. All reflections (represented as squares) followed by further reflections (represented as triangles) are grouped into a cluster C1.

[0052] In Figure 11 A lidar sensor 1 and three objects 2, 3, and 9 are shown, where object 3 is transmissive, for example, a window 7 of a vehicle 6. All reflections (represented as squares) followed by further reflections (represented as triangles) are grouped into clusters C1 and C2. Cluster C2 has a planar surface and is therefore marked. For all laser beams passing through the volume defined by cluster C2, corresponding reflections are marked as false positives E.

Claims

1. Method for evaluating reflections of lidar pulses (R1 to R4) of a lidar sensor (1) - the lidar pulses (R1 to R4) being sent by the lidar sensor (1) in different directions of a scanning region of the lidar sensor (1) in the course of a scanning operation of an environment and reflections of the lidar pulses (R1 to R4) being detected by the lidar sensor (1), - and clusters (C1, C2) being generated by clustering the reflections, characterized in that - the environment is a vehicle environment, - for each laser pulse (R1 to R4) which is reflected back to the lidar sensor (1) in the scanning region thereof, it is checked whether it is reflected back multiple times at different distances, - for the multiple reflected-back laser pulses (R1 to R4), the reflections received first in each case are determined as the first reflections in each case, - the clustering is carried out using the first reflections and - when a distance evaluation of reflections from a generated cluster (C1, C2) shows that the laser pulses (R1 to R4) are reflected there at a reflecting surface which is at least approximately homogeneous, subsequent reflections following the first reflection in each case are marked as false-positive detections (E).

2. Method according to claim 2, characterized in that a main component analysis is carried out to determine the surface of the clusters (C1, C2), a distinction being made between - clusters (C1, C2) with a planar surface, - clusters (C1, C2) with a curved surface, - clusters (C1, C2) which extend in exactly one spatial direction, and - clusters (C1, C2) which extend in three spatial directions.

3. Method according to claim 2, characterized in that only clusters (C1, B2) with a planar surface and clusters (C1, C2) with a slightly curved surface, of which the curvature corresponds to the curvature of a window panel of a vehicle, are characterized as approximately homogeneous reflective surfaces.

4. Method according to claim 2 or claim 3, characterized in that clusters (C1, C2) which extend in exactly one spatial direction are characterized as clusters (C1, C2) which represent edges of objects (2, 3, 9, 10).

5. Method according to any of claims 2 to 4, characterized in that clusters (C1, C2) which extend in three spatial directions are characterized as clusters (C1, C2) which represent dust and / or fog and / or fine-grained structures.

6. Device for evaluating reflections of lidar pulses (R1 to R4), - the device having a lidar sensor (1) which is configured to send the lidar pulses (R1 to R4) in different directions of a scanning region of the lidar sensor (1) in the course of a scanning operation of an environment and to detect reflections of the laser pulses - and an evaluation unit being provided which is designed to generate clusters (C1, C2) by clustering the reflections, characterized in that the environment is a vehicle environment and the evaluation unit is further designed - for each laser pulse (R1 to R4) which is reflected back to the lidar sensor (1) in the scanning region, to check whether it is reflected back multiple times at different distances, - for the multiple reflected-back laser pulses (R1 to R4), to determine the reflections received first in each case as the first reflections in each case, - to carry out the clustering using the first reflections, - when a distance evaluation of reflections from the generated cluster (C1, C2) shows that the laser pulses (R1 to R4) are reflected there at a reflecting surface which is at least approximately homogeneous, to mark subsequent reflections following the first reflection in each case as false-positive detections (E).