DE102018008904A1 - Method for locating a vehicle in a high definition map based on landmark objects and sensor fusion module for such a method - Google Patents

Abstract

The invention relates to a method for localizing a vehicle moving along a travel path (3) on a high definition (HD) map, wherein a geoposition of the vehicle is determined, a potential landmark object (2.2 ') by means of an environment detection module (1.1) in a vehicle coordinate system ( 4) and registered with at least one landmark object (2.2) of the High Definition (HD) map and from this a more accurate location of the vehicle than the geoposition is determined. In the method, a reference line (3.1) arranged along the travel path (3) is determined, the potential landmark object (2.2 ') and the at least one landmark object (2.2) are projected onto the reference line (3.1) and registered along the reference line (3.1). The invention further relates to a sensor fusion module (1.2 ') for the fusion of data from environmental sensors of an environment detection module (1.1) for such a method.

Description

The invention relates to a method for locating a vehicle according to the preamble of claim 1. The invention further relates to a sensor fusion module for carrying out such a method.

Automated driving (AD) methods of a vehicle rely heavily on high definition maps (HD maps). In order to use HD cards for an AD procedure, the position of a vehicle on the HD card must be determined. For georeferenced HD maps with assigned geopositions, a position determination by means of a global navigation satellite system (GNS system) is possible. To improve the accuracy against such a position determination by means of a GNS system, methods known from the prior art additionally use stationary landmarks whose respective geoposition along the travel path is known.

Such landmarks may be formed, for example, as a traffic signal, posts or traffic sign bridge. However, landmarks based on more abstract features are also possible. Landmarks can be detected with environmental sensors of a vehicle. Environmental sensors can be designed, for example, as a camera, as a radar sensor or as a light detection and ranging (LIDAR) sensor.

In the position determination method known from the prior art, the position of a vehicle in a georeferenced HD map is approximately determined by means of a GNS system and the road segment associated with the approximately determined position, including the landmarks arranged along this road segment, is provided. Environmental sensors of the vehicle detect the vehicle environment. Data from various environmental sensors and / or data determined at various locations along the travel path of a vehicle are merged, tracked and filtered by a Sensor Fusion method.

Landmarks acquired from environmental sensor data are matched against the landmarks recorded on the HD map in the associated road segment. From this comparison, the position of the vehicle in the HD card is determined.

The position of the objects detected by the environmental sensors is indicated in a vehicle coordinate system related to the vehicle. The position of the landmarks recorded in the HD map is indicated in a map coordinate system, which is usually aligned in the direction of travel of the vehicle. The map coordinate system has anchor points to which geopositions are stored.

A position hypothesis is determined from a comparison of an object detected using the environmental sensors with a landmark recorded in the HD card. Since an object is generally compared to multiple landmarks and / or a plurality of objects is detected, a plurality of position hypotheses are determined.

From the majority of the position hypotheses, a position on the HD map is determined by means of a probabilistic filter, which position is indicated in the map coordinate system.

The document DE 10 2016 201 249 A1 describes a device and a method for determining a roadway model, are compared with the sensory detected explicit information about a lane course with other, implicit information about the lane course. If the explicit information is consistent with the implicit information about the lane course, then the lane model can be calculated based on the explicit information. If a discrepancy is found between the explicit and the implicit information, the calculation of the roadway model is based on the implicit information. In this way, a particularly efficient, reliable and robust calculation of a roadway model for the lanes of a vehicle can be determined.

It is an object of the present invention to provide an improved method for locating a vehicle on a high definition (HD) map and an improved sensor fusion module for merging data from environmental sensors to perform such a method.

With regard to the method, the object is achieved by the features of claim 1. With regard to the sensor fusion module, the object is achieved by the features of claim 3.

Advantageous embodiments of the invention are the subject of the dependent claims.

In a method for locating a vehicle moving along a travel path on a high definition (HD) map, a geoposition of the vehicle is determined, a potential landmark object is detected by means of an environment detection module in a vehicle coordinate system and at least one landmark object of the vehicle Registered High Definition (HD) map and from this a geo position more accurate localization of the vehicle determined. According to the invention, a reference line arranged along the travel path is determined. The potential landmark object and the at least one landmark object are projected onto the reference line and registered along the reference line.

An advantage of the method is that registration of the potential landmark object described by the HD card occurs along only one coordinate. As a result, computing effort can be saved and at the same time the reliability and robustness of the method can be improved.

Embodiments of the invention are explained in more detail below with reference to drawings.

• 1 schematisch eine Sensorfusionsschicht nach dem Stand der Technik,
• 2 schematisch eine Sensorfusionsschicht gemäß einer Ausführungsform der Erfindung,
• 3A schematisch eine Straße mit einer Referenzlinie,
• 3B schematisch eine Straße mit einer Referenzlinie und Elementen der Straßenausstattung,
• 4A, 4B schematisch die Entwicklung von kartenbasierten Nebenbedingungen,
• 5 schematisch die Projektion eines potenziellen Landmarkenobjekts auf eine Referenzlinie einer Straße sowie
• 6 schematisch den Ablauf einer Sensorfusion für ein Landmarkenobjekt.

Showing:

• 1 schematically a sensor fusion layer according to the prior art,
• 2 schematically a sensor fusion layer according to an embodiment of the invention,
• 3A schematically a road with a reference line,
• 3B schematically a road with a reference line and elements of street equipment,
• 4A . 4B schematically the development of card-based constraints,
• 5 schematically the projection of a potential landmark object on a reference line of a road as well
• 6 schematically the flow of a sensor fusion for a landmark object.

Corresponding parts are provided in all figures with the same reference numerals.

1 schematically shows the architecture of a sensor fusion layer 1 comprising an environment detection module 1.1 a vehicle, not shown, a sensor fusion module 1.2 and an object detection module 1.3 According to the state of the art. The sensor fusion module 1.2 processes data from heterogeneous environmental sensors of the environmental detection module 1.1 and translates this data into a unified and single data view, in this case a model, all by means of the environment sensors of the environment detection module 1.1 describes perceived objects.

Fields of view of several environmental sensors are spatially and semantically combined. Where fields of view of heterogeneous environmental sensors overlap, an improved detection rate, for example, higher sensitivity and / or higher specificity and / or reduced false positive rate may be achieved. For example, different types of environmental sensors can be combined. Thus, environmental camera sensors are well-suited for capturing contextual information that describes the similarity of a captured image to an object class or a landmark prototype. Furthermore, environmental sensors configured as cameras are well suited for measuring a height, that is to say an extent perpendicular to the roadway plane. On the other hand, environment sensors designed as a LIDAR or a radar are well suited for the two-dimensional detection of an object along a plane parallel to the roadway plane.

The sensor fusion module 1.2 includes a dynamic sensor fusion submodule 1.2.1 and a static sensor fusion sub-module operating separately therefrom 1.2.2 , each with data from the environment sensors of the environment detection module 1.1 be fed.

The dynamic sensor fusion submodule 1.2.1 identifies and parameterizes dynamic objects 2.1 , The static sensor fusion submodule 1.2.2 identifies and parameterizes landmark objects 2.2 as well as other static objects 2.3 ,

2 schematically shows a sensor fusion layer 1' according to an embodiment of the invention. Contrary to the prior art, the sensor fusion layer comprises 1' According to the invention, a prepositioning module 1.4 , which contains data from the environment sensors of the environment detection module 1.1 as well as with mapping data 1.5 is fed. The mapping data 1.5 include data describing at least a portion of the travel path of the vehicle. Such mapping data 1.5 can be taken from an HD card.

The prepositioning module 1.4 created based on the environmental sensors data of the environment detection module 1.1 and the mapping data 1.5 card-based constraints 1.6 which describe restrictions on the landmarks to be found. The card-based constraints 1.6 can describe what kind of landmark is to be found. For example, the card-based constraints 1.6 then decide whether to find a traffic signal or a traffic sign bridge.

Furthermore, the card-based constraints 1.6 the position of one or more Landmarks and / or trajectories along which landmarks appear.

The sensor fusion layer 1' includes a sensor fusion module 1.2 ' , which is modified from the prior art in that it is in addition to the dynamic sensor fusion submodule 1.2.1 a generic static sensor fusion submodule 1.2.2 ' and a static landmark fusion submodule 1.2.3 includes.

The landmark fusion submodule 1.2.3 generates landmark objects 2.2 , The generic static sensor fusion submodule 1.2.2 ' generates static objects 2.3 with the exception of landmark objects 2.2 ,

Requirements regarding the recognition of landmarks as well as properties of landmark objects 2.2 differ significantly from properties of other, generic static objects 2.3 , which are usually detected for collision avoidance and to determine a free maneuvering space. Such generic static objects 2.3 For example, they may be formed as a charge lost by other vehicles, standing other vehicles, or road obstacles or barriers with which the vehicle may potentially collide.

The sensor fusion layer 1' thus separates methods for the recognition of generic static objects 2.3 of methods for recognizing landmark objects 2.2 , An advantage of this separation of recognition methods is that for the recognition of landmark objects 2.2 characteristic features are usable, which, however, for the detection of generic static objects 2.3 not available or not useful.

For example, landmarks are objects 2.2 often accessible or otherwise arranged outside the travel path of a vehicle, such as the side of the roadside or next to a guardrail. By contrast, collision avoidance involves the recognition of generic static objects 2.3 along the driveway important and can be on the detection of the vehicle closest to the static object 2.3 be limited.

Landmark objects 2.2 are often arranged on a traffic sign bridge or behind a junction or turn to the guideway. Generic static objects 2.3 on the other hand can be arranged arbitrarily.

In addition, the position and the type or the characteristics of landmark objects 2.2 recorded in an HD card and thus known. In contrast, the position, size and shape of generic static objects 2.3 generally not known.

Furthermore, landmark objects 2.2 preferably used for the longitudinal positioning, that is: for the position determination along the guideway. Therefore, it may be sufficient to keep the distance along the driveway to a landmark object 2.2 to determine. By contrast, for the detection of generic static objects 2.3 In general, a two-dimensional localization, that is: the recognition of both x-coordinates and y-coordinates important.

False-positive detections as well as false-negative detections of landmark objects 2.2 do not affect the accuracy of a localization, or only slightly. In contrast, both false-positive detections and false-negative detections of generic static objects 2.3 Significant negative effects, for example, by an unfounded emergency braking due to a false-positive detected obstacle is triggered or fails an emergency to collision avoidance with an unrecognized obstacle.

The requirements for landmark objects also differ with regard to the desired time of a recognition 2.2 of which generic static objects 2.3 , For landmark objects 2.2 a detection at the time of passing is sufficient. For generic static objects 2.3 is, for example, for timely triggering a braking or Umfahrungsmanövers, however, the earliest possible detection desirable or even required.

Due to the different requirements for recognition of landmark objects as shown 2.2 and generic static objects 2.3 may be due to a separation of the signal and processing path established by the landmark fusion submodule 1.2.3 or by the generic static sensor fusion submodule 1.2.2 ' runs, computing time can be saved and it can the recognition quality for landmark objects 2.2 and / or generic static objects 2.3 be improved.

Use an HD card to determine card-based constraints 1.6 an at least partial prepositioning is required with which a lateral or lateral position is determined with a deviation of at most two meters, preferably with a deviation of at most one meter. The lateral or lateral position can be determined by comparing environmental sensors, for example by means of at least one camera Lane markings having lane markings mapped in the HD map and / or by measuring the lateral distance from the vehicle to guardrails, for example by means of side radar, and / or by observing trajectories of other vehicles and registering those trajectories with lanes mapped in the HD map are, and / or by a sufficiently accurate geoposition by satellite navigation.

Further, prepositioning may be the determination of a tilt angle α of the vehicle relative to the course of the road 3 include. The angle of inclination α can be determined by an orientation along lane markings, which are detected optically, for example by means of a camera, and / or by orientation along guardrails and / or by determining the orientation of a vehicle trajectory determined from a sequence of geopositions, provided that the geopositions have a have sufficient accuracy.

3A schematically shows an embodiment for the determination of card-based constraints 1.6 while driving along a road 3 , The street 3 is as a reference line 3.1 modeled, which in this case coincides with the road boundary leftmost relative to the direction of travel. The reference line 3.1 , the geometry of the road 3 as well as the longitudinal offset and the arrangement of landmark objects 2.2 along the reference line 3.1 are removable from a HD card. In this case, the longitudinal offset designates the position of a projection of the landmark objects 2.2 on the reference line 3.1 , Here is a landmark objects 2.2 designed as a traffic sign, with a lateral offset S1 relative to the reference line 3.1 at the edge of the street 3 is arranged.

Optionally, an HD card can have a lateral offset S2 to S6 other elements of street equipment of the street 3 specify as in 3B shown schematically.

The 4A and 4B show schematically the development of card-based constraints 1.6 , 4A shows a section of an HD map with a road 3 , 4B shows the superimposition of a vehicle coordinate system 4 over the HD map section of the 4A , The vehicle coordinate system 4 has an x-coordinate 4.x and a y-coordinate 4.y on. The x-coordinate 4.x has approximately in the direction of travel of a not shown, the coordinate origin 4.o of the vehicle coordinate system 4 located vehicle. The y-coordinate 4.y is perpendicular to the x-coordinate 4.x , The coordinates 4.x . 4.y span a plane in which the road 3 lies.

The origin of the coordinates 4.o has a lateral offset S to the reference line 3.1 on. The x-coordinate 4.x can be at a tilt angle α opposite the course of the reference line 3.1 be inclined.

5 schematically shows a model of the road 3 , The model includes limitations 5 each with a left-sided edge point 5.L and a right-sided edge point 5.R , Between the edge points 5.L . 5.R is a border width 5.S that the reference line 3.1 perpendicular in a longitudinal position 5.P cuts. The longitudinal position 5.P is by the distance from a starting point along the unrolled reference line 3.1 determined as the projection of the coordinate origin 4.o on the reference line 3.1 given, the reference line 3.1 in a lateral distance 3.1.S from the coordinate origin 4.o lies. That by the totality of all limits 5 and the reference line 3.1 given model of the road 3 is in the vehicle coordinate system 4 aligned and displayed.

From the sensor data is also a potential landmark object 2.2 ' at one in the vehicle coordinate system 4 given certain position. Such a potential landmark object 2.2 ' For example, it may have been determined from radar data that has little or no contextual information about the shape, extent, or type of the potential landmark object 2.2 ' provide.

By aligning the model of the road 3 relative to the vehicle coordinate system 4 it is possible, by means of interpolation, to project a projection of the potential landmark object 2.2 ' on the reference line 3.1 at a longitudinal position 2.2 ' .P and in a lateral distance 2.2 ' To determine .S. Thus, an alignment of the potential landmark object 2.2 ' with a landmark object actually recorded in the HD map 2.2 possible. From this adjustment can be a correction of the longitudinal position 2.2 ' .P are determined. This allows for improved accuracy of localization of the vehicle.

6 schematically shows the flow of a sensor fusion for a landmark object 2.2 , The method processes data describing the vehicle environment from an environment detection module 1.1 and geoposition data obtained from a geoposition module 1.7 to be provided. The geoposition module 1.7 processes data from a Global Navigation Satellite System ( GNS ).

By means of a card registration function 6 become HD map data 7 brought an HD map in coincidence with the geoposition of the vehicle and from it mapping data 1.5 that determines the Course of the driveway 3 relative to a model of the road 3 describe.

The mapping data 1.5 become a prepositioning function 8th continues to slide. The prepositioning function 8th determines from this as well as from the data of the environment detection module 1.1 card-based constraints 1.6 , The prepositioning function 8th further determines an estimate for the lateral position and orientation of the vehicle relative to the road 3 , In other words: the prepositioning function 8th determines the projection and the angle of inclination α of the vehicle coordinate system 4 relative to the reference line 3.1 the street 3 ,

Further, from the mapping data 1.5 Landmark positions 9 determines which the positions of landmark objects 2.2 relative to the vehicle coordinate system 4 and / or relative to the coordinate system of the HD map data 7 specify.

The method further includes a landmark tracking function 10 from the data of the environment detection module 1.1 as well as from the card-based constraints 1.6 potential landmarks 2.2 ' determined.

A localization feature 11 determined from the potential landmark objects 2.2 ' and the compared landmark positions 9 a vehicle position 12 , which indicates the position of the vehicle in an HD map. In other words, by matching the landmark positions 9 with the potential landmark objects 2.2 ' determines the localization function 11 the exact longitudinal position of the vehicle along the reference line 3.1 , In addition, the localization feature 11 determine or refine further parameters that determine the location of the vehicle relative to the HD map data 7 describe.

An advantage of the method according to the invention is that landmark objects 2.2 along a one-dimensional coordinate, which is the distance along the reference line 3.1 the street 3 can be determined, tracked and recognized.

Objects that are both landmark objects 2.2 as well as generic static objects 2.3 or dynamic objects 2.1 can be on the reference line 3.1 based. As a result, at least one coordinate transformation can be saved compared to previously known methods. The lateral or lateral offset S an object opposite the reference line 3.1 is used as a feature of an object descriptor.

Thus, calculation time for the sensor fusion and for the determination of the vehicle position 12 be saved.

In addition, the robustness and reliability of the method can be increased, since particularly test-sharp error detection methods can be used, their use for landmark objects 2.2 that are detected along multiple dimensions would be too computationally expensive.

Furthermore, the performance of determining the vehicle position 12 be improved because of the on the vehicle coordinate system 4 related HD map data 7 Location and / or type of landmarks 2.2 predictable and thus the misrecognitions of landmark objects 2.2 avoided or reduced.

In one embodiment of the invention, one of the HD map data 7 determined landmark object 2.2 to the environment detection module 1.1 transmitted, which controls the at least one environment sensor such that he targeted to the position and / or the type of the transmitted landmark object 2.2 examined. This can improve the object recognition and false detection can be avoided.

LIST OF REFERENCE NUMBERS

1, 1 '

Sensor fusion layer

1.1

Area Learning Module

1.2

Sensor fusion module

1.2 '

Sensor fusion module

1.2.1

dynamic sensor fusion submodule

1.2.2

static sensor fusion submodule

1.2.2 '

generic static sensor fusion submodule

1.2.3

Landmarkenfusionssubmodul

1.3

Object detection module

1.4

Präpositionierungsmodul

1.5

mapping data

1.6

card-based constraints

1.7

Geopositionsmodul

2.1

dynamic object

2.2

Landmark Property

2.2 '

potential landmark object

2.2'.P

longitudinal position

2.2'.S

lateral distance

2.3

static object

3

Road, driveway

3.1

reference line

3.1.S

lateral distance

4

Vehicle coordinate system

4.x

x-coordinate

4.y

y coordinate

4.o

origin

5

limit

5.L

left-sided edge point

5.R

right-sided edge point

5.S

Border width

5.P

longitudinal position

6

Card registration function

7

HD map data

8th

Präpositionierungsfunktion

9

Landmark positions

10

Landmark tracking function

11

localization function

12

vehicle position

α

tilt angle

S, S1 to S6

lateral offset

QUOTES INCLUDE IN THE DESCRIPTION

This list of the documents listed by the applicant has been generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Cited patent literature

• DE 102016201249 A1 [0009]

Claims (3)

Method for locating a vehicle moving along a travel path (3) on a high definition (HD) map, wherein a geoposition of the vehicle is determined, a potential landmark object (2.2 ') detected by means of an environment detection module (1.1) in a vehicle coordinate system (4) and is registered with at least one landmark object (2.2) of the high definition (HD) map and from this a more accurate location of the vehicle compared to the geoposition is determined, characterized in that a along the travel path (3) arranged reference line (3.1) is determined, the potential Landmark object (2.2 ') and the at least one landmark object (2.2) projected on the reference line (3.1) and registered along the reference line (3.1). Method according to Claim 1 characterized in that the potential landmark object (2.2 ') and the at least one landmark object (2.2) based on a longitudinal position (2.2'.P) and on the basis of a lateral distance (2.2'.S) of the potential landmark object (2.2') relative to the reference line (3.1). Sensor fusion module (1.2 ') for the fusion of data from environmental sensors of an environment detection module (1.1) for a method according to Claim 1 or 2 , characterized in that the sensor fusion module (1.2 ') comprises a generic static sensor fusion sub-module (1.2.2') and a landmark fusion sub-module (1.2.3) decoupled therefrom, the generic static sensor fusion sub-module (1.2.2 ') for detecting static objects (1.2.2'). 2.3) on the basis of data of the environment detection module (1.1) is formed and wherein the Landmarkenfusionsssubmodul (1.2.3) for recognizing landmark objects (2.2) based on data of the environment detection module (1.1) and on the basis of card-based constraints (1.6) is formed.

Images