DE102021209403A1 - Method for generating a static environment model - Google Patents

Abstract

The invention relates to a method for generating a static environment model (5) with the following steps:
- Recording (S1) of a vehicle environment by means of at least one environment detection sensor;
- generating (S2) at least one environment representation;
- determining (S3) at least one shape point (F) in the at least one environment representation;
- Generating (S4) at least one first polyline (3);
- Entering (S5) the at least one first polyline (3) in a memory as an entered polyline;
- Renewed recording (S6) of the vehicle environment by means of at least one environment detection sensor;
- generating (S7) a further environment representation;
- Determine (S8) at least one further shape point (Fa);
- Generation (S9) of a further polyline based on the at least one further shape point (Fa)
- determining (S10) a distance between the at least one further polyline and the entered polyline;
- Adjusting (S11) the entered polyline based on the distance between the further polyline and the entered polyline;
- Generating (S12) a static environment model (5) based on the registered polyline (3).

Description

The invention relates to a method for generating a static environment model.

• - Eine Grid-Map mit Merkmalsextraktoren:
  • Eine Grid-Map sammelt die Abtastpunkte in einer diskretisierten Ebene oder in einem diskretisierten Raum und basierend auf der Belegungswahrscheinlichkeit jeder Zelle wird die Zelle als Teil des statischen Umfelds betrachtet. Um high-level-Merkmale aus der Grid-Map zu extrahieren werden Merkmalsextraktionsverfahren wie Kantendetektion, Kurvendetektion etc. benötigt.
• - Box-Tracking (basierend auf Kalman Filter) Box-Tracking ist ein wohlbekanntes Verfahren zum Objekttracking, welches auch verwendet werden kann, um Segmente des statischen Umfelds basierend auf Kalman Filtern zu tracken.
• - Stitchen von Polylinien von Sensoren zu verschiedenen Zeitstempeln
• - Tracken von Formpunkten der Liniensegmente (basierend auf Kalman Filtern)

A detection of the static environment and a representation in a simple format is an essential perception function, which is required by driving functions for autonomous driving and also for driver assistance systems. Objects (boxes) and polylines are mainly used for the display. In the case of noisy and incomplete data, a fusion and tracking of the sensor data is necessary to achieve a stable representation of the static environment. The fusion/tracking can be performed using one of the following methods.

• - A grid map with feature extractors:
  • A grid map collects the sampling points in a discretized plane or space and based on the occupancy probability of each cell, the cell is considered part of the static environment. In order to extract high-level features from the grid map, feature extraction methods such as edge detection, curve detection, etc. are required.
• - Box tracking (based on Kalman filters) Box tracking is a well-known method for object tracking, which can also be used to track segments of the static environment based on Kalman filters.
• - Stitch polylines from sensors to different timestamps
• - Tracking shape points of line segments (based on Kalman filters)

• - Grid-Map:
  • Abhängig von der Auflösung und der Größe der Karte nimmt die Komplexität zum Aufrechterhalten der Wahrscheinlichkeit der Zellen und auch der Merkmalsextraktion exponentiell zu, so dass es zu einer Herausforderung wird, ein Grid für eingebettete Systeme zu verwenden.
• - Box-Tracking:
  • Der Nachteil der Verwendung von Boxen ist, dass die Darstellung von komplexen Formen des statischen Umfelds viele kleine Boxen benötigt, welche das Tracken zeitaufwändig machen, insbesondere aufgrund der Verwendung von Kalman Filtern. Zusätzlich wird die Kontinuität der Objektgrenzen normalerweise nicht aufrechterhalten. Das bedeutet, dass die Boxen individuell aktualisiert werden, ohne irgendwelche Kopplungszwänge zischen diesen zu berücksichtigen.
• - Polylinienstitching:
  • Polylinienstitching kann verrauschte Daten nicht verarbeiten und erzeugt eine verrauschte Darstellung des statischen Umfelds.
• - Tracking von Formpunkten:
  • Dieses Verfahren hat folgende Hauptnachteile. Zum einen wie die Formpunkte ausgewählt werden sollen. Zum anderen wird jeder Formpunkt individuell aktualisiert und die Kopplungszwänge werden nicht berücksichtigt. Aufgrund der genannten Probleme ist dieses Verfahren hauptsächlich dazu nützlich, einfache Straßenbegrenzungen mit einer geringen Krümmung zu tracken.

The current tracking methods have the following problems:

• - Grid map:
  • Depending on the resolution and the size of the map, the complexity of maintaining the probability of the cells and also the feature extraction increases exponentially, making it a challenge to use a grid for embedded systems.
• - Box tracking:
  • The disadvantage of using boxes is that the representation of complex shapes of the static environment requires many small boxes, which makes tracking time consuming, especially due to the use of Kalman filters. In addition, the continuity of the object boundaries is not usually maintained. This means that the boxes are updated individually without taking into account any coupling constraints between them.
• - Polyline stitching:
  • Polyline stitching cannot process noisy data and produces a noisy representation of the static environment.
• - Tracking shape points:
  • This method has the following main disadvantages. On the one hand how the shape points should be selected. On the other hand, each shape point is updated individually and the coupling constraints are not taken into account. Due to the problems mentioned, this method is mainly useful for tracking simple road boundaries with a small curvature.

It is therefore an object of the present invention to provide a method by means of which a static environment can be generated reliably and with little computing effort.

This object is solved by the subject matter of independent claim 1. Further advantageous refinements and embodiments are the subject matter of the dependent claims.

• - Aufzeichnen des Fahrzeugumfelds mittels zumindest einem Umfelderfassungssensor;
• - Erzeugen zumindest einer Umfeldrepräsentation;
• - Ermitteln von zumindest einem Formpunkt in der zumindest einen Umfeldrepräsentation;
• - Erzeugen zumindest einer ersten Polylinie;
• - Eintragen der zumindest einen ersten Polylinie in einen Speicher als eingetragene Polylinie;
• - Erneutes Aufzeichnen (S6) des Fahrzeugumfelds mittels zumindest einem Umfelderfassungssensor;
• - Erzeugen (S7) einer weiteren Umfeldrepräsentation;
• - Ermitteln zumindest eines weiteren Formpunkts;
• - Erzeugen einer weiteren Polylinie basierend auf dem zumindest einen weiteren Formpunkt;
• - Ermitteln eines Abstandes zwischen dem zumindest einen weiteren Polylinie und der eingetragenen Polylinie;
• - Anpassen der eingetragenen Polylinie basierend auf dem Abstand zwischen der weiteren Polylinie und der eingetragenen Polylinie;
• - Erzeugen einer statischen Umfeldmodells basierend auf der eingetragenen Polylinie.

According to the invention, a method for generating a static environment model is proposed with the following steps:

• - Recording the vehicle environment by means of at least one environment detection sensor;
• - generating at least one environment representation;
• - determining at least one shape point in the at least one environment representation;
• - generating at least one first polyline;
• - Entering the at least one first polyline in a memory as entered polyline;
• - Renewed recording (S6) of the vehicle environment by means of at least one environment detection sensor;
• - generating (S7) a further environment representation;
• - determining at least one further shape point;
• - generating a further polyline based on the at least one further shape point;
• - determining a distance between the at least one further polyline and the entered polyline;
• - Adjusting the entered polyline based on the distance between the further polyline and the entered polyline;
• - Creation of a static environment model based on the entered polyline.

The at least one first polyline can be described by sorted shape points. Each shape point can be connected to a maximum of two other shape points using edges/line segments. Each shape point has the following attributes. Position in x, y coordinates and covariance matrix in x, y coordinates. The environment detection sensor is particularly preferably a radar sensor. Alternatively or cumulatively, it would also be conceivable to use a lidar or camera sensor. However, the method is independent of a specific sensor type and can be carried out with any environment detection sensor. The environment representation is, for example, a 2D point cloud that describes the visible boundaries of objects. This point cloud can be generated based on various sensor technologies and is not limited to radar sensors.

The first polyline therefore serves as a starting point and is tracked/continued through iterations of certain process steps. The determined shape points for the first polyline are determined in the first detection cycle of the at least one environment detection sensor. It would be conceivable that not just a first polyline, but several, in particular two, first polylines are generated, since the polylines are formed by the shape points on static objects and can therefore be present at least on both sides of a vehicle. Depending on the range of the sensor used or depending on the prevailing scenario, more than two polylines can also be formed, for example in an intersection area in which more than two lane boundaries are visible to the sensor. However, the shape points and the polylines formed from them not only describe lane boundaries, but can also be formed, for example, on parked vehicles or other static objects in the vehicle environment, such as house walls. The shape points are formed from the reflection points of the environment detection sensor, with not every reflection point corresponding to a shape point. Rather, several reflection points can be represented by a line segment, which is described by two shape points. The other shape points are determined accordingly in the other detection cycles of the environment detection sensor, for example the radar sensor. In general, the first polyline or the first shape points for the first polyline are determined in the first detection and further shape points of the static environment are determined in the subsequent detections in order to expand/track the at least one first polyline if necessary. The at least one first polyline is stored accordingly in an internal memory and continued accordingly. The first polyline is entered in a memory and serves as a reference for determining the distance to the other polylines. After adjusting the entered polyline, this adjusted entered polyline is used as a reference in the further cycles. The second detection cycle thus begins after the first polyline has been entered. The procedural steps after the entry are repeated again and again in the further course in order to adapt the entered polyline and to expand the static environment model.

In a particularly preferred embodiment, the Euclidean distance between the further polyline and the entered polyline is determined. Determining the distance is advantageous in that it can be determined whether the further shape point is relevant for the at least one first polyline. If the distance is too large or cannot be determined directly, this would indicate another element of the static environment, which in turn can be described with its own polyline.

In a further preferred embodiment, the polyline entered is adjusted accordingly depending on whether the Euclidean distance exceeds or falls below a predefinable limit value. If the distance between the further polyline and the entered polyline or one of the shape points of the at least one first polyline is small, this adjustment can consist in merging the further polyline with the entered polyline. If the point is within the maximum amount and on the line segment, a line segment adjustment is performed. If the point is not on the line segment, a new line segment is created between the first polyline and the additional shape point and the additional shape point is added to the polyline. To determine the distance from the entered polyline to the other, measured polyline, first the distances between the form points of the entered polyline and the line segments of the other polyline and the distances of the form points of the other polyline to the line segments of the entered polyline are determined. The smallest distance is selected from these determined Euclidean distances and, based on this, it is determined whether the smallest distance is above or below a predetermined limit value. If the Euclidean distance is less than a specified limit value, the other measured polyline is associated with the entered polyline. Is the distance greater than the limit value, the further polyline is entered separately and is then available for the further detection cycles as a further or additional polyline entered.

Furthermore, it is preferably provided in one configuration that a large number of first polylines are generated based on a position and number of the determined shape points. The shape points determined for this are thus determined in a first detection cycle from the reflection points of the radar sensor used, for example, and connected according to their position to form more than one first polyline. Thus, in this embodiment, several first polylines would be created, which are expanded/tracked by further determined shape points in the following detection cycles of the surroundings detection sensor.

In a further embodiment of the invention, it is particularly preferable for the polyline entered to be simplified by reducing the number of shape points when a definable number of shape points is exceeded. During simplification, so-called critical shape points are selected using a recursive method and the remaining shape points are reduced. In this method, a line is considered between the first and last shape point of the polyline, and then the distance from all other points to the line is calculated. If the maximum distance is greater than a threshold, the associated point is considered a break point, dividing the polyline into two clusters. The same procedure is carried out for each new cluster until the distance between the points and the line is less than a predefinable limit value. Then these points are removed. At the end only the start point, the end point and the breakpoints remain.

In a further preferred embodiment, a simplified polyline is generated from the entered polyline by simplifying. This simplified polyline is formed from the start point, the end point and the remaining breakpoints. One breakpoint or several breakpoints can be determined, depending on the course of the polyline and depending on which limit value is set for the distance between the points. The simplification also reduces the computational effort. This is particularly advantageous for very long polylines, because without the simplification, every shape point and every line segment would always have to be taken into account/calculated as long as the vehicle is moving and the polyline(s) entered is/are extended. In this case, the static environment model would then be generated from the simplified polyline(s).

• 1: ein schematisches Ablaufdiagramm einer Ausführungsform der Erfindung;
• 2a-c: je eine schematische Darstellung von reflektierten Sensorpunkten und getrackten Polylinien;
• 3a-c: je eine schematische Darstellung einer Polylinie gemäß einer Ausgestaltung der Erfindung;
• 4: eine schematische Darstellung einer Vereinfachung der Polylinie gemäß einer Ausgestaltung der Erfindung.

Further advantageous refinements and embodiments of the invention are the subject of the drawings. Show in it:

• 1 : a schematic flow diagram of an embodiment of the invention;
• 2a-c : a schematic representation of reflected sensor points and tracked polylines;
• 3a-c : each a schematic representation of a polyline according to an embodiment of the invention;
• 4 1: a schematic representation of a simplification of the polyline according to an embodiment of the invention.

In 1 a schematic flow diagram of an embodiment of the invention is shown. In step S1, a vehicle environment is recorded using at least one environment detection sensor. At least one environment representation is generated in step S2. In the following step S3, at least one shape point is determined in this environment representation. At least one first polyline based on the at least one first shape point is then generated in step S4 using a line segment. In step S5, the at least one first polyline 3 is entered into a memory as an entered polyline. In step S6, the vehicle environment is again recorded using the at least one environment detection sensor. A further environment representation is then generated in step S7. At least one further shape point is determined in step S8. In step S9, another polyline is generated based on the at least one other shape point 2a. In the subsequent step S10, a Euclidean distance between the further polyline and the entered polyline is determined. In step S11, the entered polyline is adjusted based on the distance between the further polyline and the entered polyline. Finally, in step S12, a static environment model is generated based on the entered polyline. To determine the distance from the entered polyline to the other, measured polyline, first the distances between the form points of the entered polyline and the line segments of the other polyline and the distances of the form points of the other polyline to the line segments of the entered polyline are determined. The smallest distance is selected from these determined Euclidean distances and, based on this, it is determined whether the smallest distance is above or below a predetermined limit value. If the Euclidean distance is less than a specified limit value, the other measured polyline is associated with the entered polyline. If the distance is greater than the limit, the additional polyline is entered separately worn and is then available for the further detection cycles as a further or additional entered polyline.

Steps S1 to S5 describe the first detection cycle of the environment detection sensor, in which the first polyline is created and then entered. Steps S6 to S12 describe the second and all subsequent detection cycles.

That is, steps S6 to S12 are repeated as long as the vehicle is operated.

The 2a , 2 B and 2c each show a schematic representation of reflected sensor points and tracked polylines. The left representation of 2a FIG. 1 shows an ego vehicle 1 in a crossing scenario with a schematically illustrated detection range E of an environment detection sensor, not shown here. In this detection area E, sensor points 2 or reflection points 2 are now determined. These reflection points 2 are detected on all objects in the area surrounding the ego vehicle 1 . The form points for the polylines are calculated from these reflection points 2. At this point in time, the environment detection sensor of the ego vehicle is already in an advanced detection cycle, for example cycle 23. At this point in time, the at least one first polyline has already been determined and already expanded/tracked with the shape points shown here. In the right picture the 2a it is shown that shape points were formed from the reflection points 2, which were determined on static objects 4 in the environment of the ego vehicle 1, and polylines 3 were formed from the shape points. These polylines 3 then represent an environment model 5 of the static environment. This procedure is repeated in each detection cycle. Various advanced cycles are shown here as examples to show how an environment representation of the static environment can be reconstructed.

2 B shows how too 2a a detection cycle of the environment detection sensor, the cycle in 2 B is more advanced, which can also be seen from the advanced position of the ego vehicle 1 . Reflection points 2 are again determined here on the static objects 4, among other things. In this case, these static objects 4 are, for example, house walls, lane boundaries or parked vehicles, as shown. In the right image, the polylines of the 2a This detection cycle was extended by means of the shape points and thus the static environment model 5 was also extended. The 2c 12 now shows an even more advanced ego vehicle 1 in an even later detection cycle, in which the ego vehicle has already left the intersection scenario. In the right picture the 2c it can be seen how the polylines 3 have expanded through the detection cycles and thus also the static environment model 5 .

The 3a , 3b and 3c each show a schematic representation of a polyline according to an embodiment of the invention. These figures show how a polyline 3 can be adapted after determining at least one further shape point Fa. In this case, the polyline 3 shown here already includes three shape points F. Now a further shape point Fa is determined in the longitudinal direction and the Euclidean distance between the polyline 3 and the further shape point Fa is determined. As shown, the Euclidean distance between the further shape point Fa and the last shape point F of the polyline 3 is below a certain limit, so the further shape point Fa is merged with the last shape point F and a fitted polyline 3a is produced. A Kalman filter update rule or simple averaging can be used for the merging.

In the representation of 3b the further shape point Fa is determined in the lateral direction to the polyline 3 . The line segment of the plotted polyline 3 is updated based on fitting a new line segment to three points: two shape points F of the plotted polyline 3 and another shape point Fa of another measured polyline. The two shape points F of the polyline 3 and the line segment between these two shape points move accordingly.

In the representation of 3c the Euclidean distance between the polyline 3 and the further shape point 2a is greater than a specified limit value. Therefore, when polyline 3 is adjusted, the adjusted polyline 3a is extended with a new line segment.

The 4 shows a schematic representation of a simplification of a polyline 3 according to an embodiment of the invention. The polyline 3 in this representation comprises a plurality of shape points F. To simplify this, a straight line is now formed between the first and the last shape point and then the distance from all other shape points F to the line is calculated. If the maximum distance is greater than a threshold, the associated shape point F is considered to be a break point Fb dividing the polyline 3 into two clusters. The same procedure is carried out for each new cluster until the distance between the points F and the line is less than a predefinable limit value. Then these Points F removed. At the end only the starting point, the end point and the breakpoints Fb remain. In this illustration, the simplified polyline 3b has been simplified to a start point, an end point and a breakpoint Fb.

Reference List

1

ego vehicle

2

reflection point

3

polyline

3a

fitted polyline

3b

simplified polyline

4

static object

5

environment model

E

detection range

f

shape point

fa

another shape point

color

breakpoint

S1-S12

process steps

Claims (6)

Method for creating a static environment model (5) with the following steps: - Recording (S1) of a vehicle environment by means of at least one environment detection sensor; - generating (S2) at least one environment representation; - determining (S3) at least one shape point (F) in the at least one environment representation; - Generating (S4) at least one first polyline (3); - Entering (S5) the at least one first polyline (3) in a memory as an entered polyline; - Renewed recording (S6) of the vehicle environment by means of at least one environment detection sensor; - generating (S7) a further environment representation; - Determine (S8) at least one further shape point (Fa); - Generation (S9) of a further polyline based on the at least one further shape point (Fa) - determining (S10) a distance between the at least one further polyline and the entered polyline; - Adjusting (S11) the entered polyline based on the distance between the further polyline and the entered polyline; - Generating (S12) a static environment model (5) based on the registered polyline (3). procedure after claim 1 , characterized in that a Euclidean distance between the additional polyline and the entered polyline (3) is determined. procedure after claim 2 , characterized in that depending on whether the Euclidean distance exceeds or falls below a predefinable limit value, a corresponding adaptation of the entered polyline is carried out. Method according to one of the preceding claims, characterized in that a large number of first polylines (3) are generated based on a position and number of the determined shape points (F). Method according to one of the preceding claims, characterized in that when a definable number of shape points (F) is exceeded, the polyline entered is simplified by reducing the shape points (F). procedure after claim 5 , characterized in that a simplified polyline (3b) is generated by simplifying the entered polyline.

Images