Method for locating a position of a vehicle by detecting geometric features of the environment

Abstract

A method is proposed for localizing an at least partially automated vehicle using its environmental sensors, wherein, to carry out the localization, the vehicle is provided with an HD map with map information, at least comprising information on roadways, wherein the HD map contains information on geometric features of static objects in the vicinity of the roadways, which can be detected by any sensor type of the environmental sensors, and wherein the localization is carried out in such a way that the vehicle continuously determines its current position by carrying out an environmental detection using its environmental sensors and comparing geometric features of detected static objects at the current position assumed by the vehicle with the geometric features of the same static objects stored in the HD map, wherein in the case in which the comparison results,that for at least one geometric feature detected by the environment detection, a predetermined degree of agreement exists, the current position of the vehicle is confirmed or determined as the actual vehicle position.,

Description

The present invention relates to the field of locating vehicles, particularly in urban environments.The localization of vehicles, i.e. the determination of their current position, is important and necessary for many situations. For vehicles with an automation level of 2 or higher, i.e. partially automated or highly automated vehicles up to autonomous vehicles, highly precise localization is necessary in order to be able to operate automated functionalities such as autonomous driving.To locate the position of vehicles, satellite data is usually acquired and processed via a GNSS (global navigation satellite system). The use of satellite data for position determination is very useful in most applications. However, in the urban environment, in particular in a very densely installed urban environment, problems arise in the acquisition of the satellite signals, since buildings hide the view or the signals are reflected by surfaces, so that the precision of the position determination is significantly reduced and in some cases even no position determination is possible.One known solution to address these issues is to use sensor-based systems to determine position. Typically, a sensor-based map of a particular environment is created as a reference map and stored in the vehicle. Sensors used in this case are, for example, lidar or radar. The vehicle may then compare this map to data captured by its in-vehicle sensors and determine where it is located based on the comparison.Since there is still a need for improvement with regard to the accuracy of the localization, the object of the invention is to provide an improved method for the localization of vehicles.This object is achieved by the features of the independent claims. Advantageous embodiments are the subject of the dependent claims.A method for locating an at least partially automated driving vehicle using its environment sensor system is proposed, wherein, in order to carry out the locating, an HD map with map information is provided to the vehicle, at least comprising information on lanes, wherein the HD map contains information on geometric features of static objects in the environment of the lanes, which features can be detected by each sensor type of the environment sensor system, and wherein the locating is carried out in such a way that the vehicle continuously determines its current position by carrying out an environment detection by means of its environment sensor system and matching geometric features of detected static objects at the current position assumed by the vehicle with the geometric features of the same static objects stored in the HD map, wherein, in the case in which the matching results, that a predetermined degree of agreement is present for at least one geometric feature recognized by the environment detection, the current position of the vehicle is confirmed or determined as the actual vehicle position.In one embodiment, the geometric features of the static objects comprise at least one edge of a building or of a traffic sign, a corner of a building or of a traffic sign, a window frame, a post of a sign or of a traffic sign, a roadway marking, a curb, a barrier.In one embodiment, in the event that the degree of agreement falls below a predefined limit value, a measure is taken.In one embodiment, the measure is one or a combination of: carrying out the localization again, calibrating a position sensor of the vehicle, matching with an alternative, redundant localization, stopping the vehicle.In one embodiment, a threshold value for the predetermined degree of agreement is set in advance for predetermined lanes.In one embodiment, the geometric features of the static objects detected by the environment sensor system are divided into geometric features of the static objects close to the roadway and located above the roadway, processed, and stored in separate HD maps.In one embodiment, the use of the HD maps is weighted as a function of the current driving situation.Furthermore, a computer program is provided which is configured to execute individual steps or all steps of the method by means of program code on an in-vehicle computing unit.Furthermore, a computing unit of a vehicle is provided on which the computer program is implemented.Furthermore, an at least partially automated driving vehicle is provided, having at least one environment sensor system for detecting the environment and an in-vehicle computing unit on which the computer program is implemented.Further features and advantages of the invention are evident from the following description of exemplary embodiments of the invention, on the basis of the figures of the drawing which shows details according to the invention, and from the claims. The individual features can each be realized individually or severally in any combination in a variant of the invention.Preferred embodiments of the invention are explained in more detail below with reference to the attached drawings. FIG. 1 is a schematic illustration of localization according to an embodiment of the present invention.As already mentioned at the outset, it is often difficult to provide as precise a localization (position determination) of a vehicle 1 as possible, in particular in the densely installed urban environment, since the previous systems with satellite- or sensor-assisted position determination (or a combination thereof) are often not accurate enough in such an environment.In order to reduce the influence of GNSS on the at least partially automated driving, a so-called sensor-based localization is used in addition to the localization by means of GNSS. For this purpose, an HD map typical of the sensor used is typically generated and the vehicle 1, more precisely its internal computing unit 11, balances the real-time data recorded by the vehicle's surroundings sensors with the map data in order to determine the position of the vehicle 1. For safety reasons, not only is a single sensor type used for localization, but several different types, i.e., e.g., lidar and radar and camera. The use of different sensor types is therefore necessary and expedient since, due to the different sensor properties in the detection of objects, not every sensor can detect everything in every situation in order to be able to carry out a localization reliably. For example, the environment detection of cameras, i.e. optical sensors, and lidar is limited in the case of rain, whereas this is not the case in radar.Currently, different HD cards, also referred to as card layers, are generated, monitored and kept up-to-date for each sensor type. This requires download of a very large amount of data by each vehicle. This means that each individual map layer must also be monitored in order to be able to detect a change in the static environment reliably and precisely, since accurate and secure localization can only be ensured in the case of current HD maps. However, monitoring different card layers requires large cloud infrastructures to provide the necessary computing and storage performance. A further aspect is that different map layers can represent different features (feature) on account of different detection properties of the different sensor types. For example, due to fog, a geometric feature of a traffic light cannot be captured by a camera and is therefore removed from the corresponding map layer or is not present there. The radar, on the other hand, detects this geometric feature, since radar is not influenced by fog, so that the geometric feature of the traffic light is still present in the map layer of the radar. Thus, different information is contained in different card layers.In order to counter these problems, it is proposed to use only a single map layer in which the geometric features are present that can be detected by all sensor types. Such an HD card with only a single card layer can be effected by appropriate filtering of already existing HD cards (with a corresponding number of layers). Alternatively, an HD map can also be newly generated by performing the filtering when generating the HD map. The HD card with only one layer drastically reduces the amount of data required. In addition, the map data is available more quickly because the download size is smaller and less bandwidth is needed.By using only a single map layer, in which geometric features of an environment that can be detected by all sensor types are contained, the monitoring and detection of the geometric features becomes significantly more precise and more reliable on the basis of the different field of view (fields of view), measurement ranges and properties of the different sensor types.Since the same geometric features of the roadway and the static environment close to the roadway can be detected by the different sensor types, the monitoring and the generation of the HD map becomes more reliable and more robust with respect to weather influences and traffic situations.Moreover, no complex training chains are necessary for detecting static environment (defined objects), since only geometric features are required for locating the vehicle and for generating the HD map. The computing power for the required environment detection and the complexity of the required HD card layer are thus reduced and the validation of the detection is simplified.The geometric features are, in particular, geometric features that can be easily detected by each of the sensor types, such as edges and corners of buildings or traffic signs, window frames, posts (signs, traffic lights, traffic signs, etc.), roadway markings, a curb edge, a barrier and other significant features that can be clearly distinguished from the background by the sensors. In this case, no classification of the features takes place, as in current methods. By using simple, significant geometric features, the algorithm used for localization can be used worldwide and no country- or route-specific adaptation is necessary.The proposed method for localization is executed and implemented by at least partially automated vehicles 1, more precisely an internal computing unit 11 thereof, since it is executed as a computer program.In FIG. 1, a localization method according to an embodiment is schematically illustrated and described below. For locating a vehicle 1, a high-resolution map, also referred to as an HD map, is provided to the vehicle. This contains all data important for highly automated driving, in particular data on roads such as roads and on available static features 2.1-2.3. Static features are, for example, (roadway) markings of all types (zebra stripes, turning arrows, lane markings, etc.) and / or traffic signs (signs, traffic lights, directions, road signs, buildings, etc.), as already mentioned. It may also contain information about concealed objects, i.e. objects which cannot be recognized by the vehicle sensor system.Information on available geometric features 2.1-2.3 of the static environment including the roadway is collected in advance for creating the HD map, e.g., by capturing by means of corresponding sensor systems and / or by using available information. As already mentioned, in contrast to the prior art, only a single map layer is present, which contains geometric features that can be detected by each of the surroundings sensors of the vehicle 1.The HD map is provided to the vehicle 1, for example, by being downloaded by the in-vehicle computing unit 11 (also referred to as a controller). In this case, either the entire HD card can be provided, or only a part of the HD card is provided. The part can be, for example, the HD map for a continent or for a country or for a region, or only the part of the HD map required for the current route. By downloading only portions of an entire HD card, both the data traffic and the storage space required to store the HD card are significantly reduced.An already existing HD map is also generally updated by the vehicle 1 with its environment sensor system 10 (the detection range 101 is shown in FIG. 1 ), wherein the detected data and information are transmitted to an external processing device 3, in which the HD map is then newly created or updated using the data / information.To locate the vehicle 1, i.e. its current position in the HD map determined by a predefined method, the roadway and the environment close to the roadway are detected by means of vehicle-mounted sensors, the so-called environment sensor system 10. In this case, geometric features 2.1-2.3 of the static environment, which are stored in the HD map and are recognizable by all sensor types, are also required for localization. During the travel of the vehicle 1, the environment is detected in real time by means of the environment sensor system 10. The detected geometric features 2.1-2.3 are matched to the geometric features 2.1-2.3 stored in the HD map, and in the case of a match to at least one predefined extent, e.g. a predefined percentage of a match, the position is determined as the actual vehicle position. This is generally done by confirming the position stored in the HD map as the current position or by transmitting the ascertained position to the vehicle 1 (e.g. as GPS coordinates).Note that the vehicle 1 always has an assumption on which position (global position, GPS position, position on a map) it is at. The geometric features 2.1-2.3 detected by the environment sensor system 10 are assigned to the assumed (assumed, assumed, or "assisted") position in order to determine the corresponding features in the HD map. The comparison with the HD map reveals whether the assumption is correct, that is to say the vehicle 1 is located at the assumed current position, or whether there are deviations from the assumption. In the event that there is a deviation, it is determined which type it is. This means, for example, whether one or more geometric features 2.1-2.3 have not been detected or have been detected at another position, so that only a certain percentage matches between real-time data and data of the HD card. If this degree of agreement, i.e. for example the percentage, does not fall below a predefined value, it can be assumed that the position assumed by the vehicle 1 corresponds to the actual position. If the degree of agreement is undershot, it is expedient to take a measure, depending on the driving situation. Such a measure can be that the localization is carried out again. A comparison with an alternative, redundant localization can also be carried out. However, it may also be that the vehicle carries out an emergency stop in order to calibrate / start its system anew.The method of localization can be extended and more precisely by dividing the detection of the geometric features 2.1-2.3 into features 2.1-2.3 close to the roadway and located above the roadway. The features 2.1-2.3 are divided directly after detection by the sensor or sensors of the environment sensor system 10 of the vehicle 1. the divided features 2.1-2.3 are then processed separately and two maps are generated, one with features 2.2 close to the roadway and one with features 2.1, 2.3 located above the roadway. By sharing features 2.1-2.3, false positive detections can be reduced. In addition, depending on the environment, a weighting can be carried out which of the maps is used more strongly for the localization. For example, a map close to the roadway for a roadway in the urban environment with few markings that are not visible or are difficult to recognize is given a less high weighting than the map with the features located above the roadway, since it is more likely, in particular in the urban environment, that building edges and road signs / traffic lights are present that can be easily recognized by all sensor types. On freeways, on the other hand, it may be exactly the other way round, since it is more likely here that the roadway markings are best recognizable.The selection of which of the maps should be weighted more strongly can be made directly during the creation of the HD map (manually or with computer assistance or only by the computing unit 11) or also during the use of the HD map by a self-learning system.The method may be implemented as a computer program that is executed on the in-vehicle computing unit 11. Further, the in-vehicle computing unit 11 is also configured to communicate with an external processing device 3 to transmit data and information thereto. In addition, it is advantageously configured to receive data and / or information from an external processing device 3, e.g. the HD card. The method is used in at least partially automated driving vehicles 1.It is advantageous in the method according to the invention for locating a vehicle 1 that the existing infrastructure of the vehicle 1 (environment sensor system 10, computing unit 11 can be used to carry out the method and to communicate externally with cloud, etc.). Thus, no provision of new hardware is necessary. In addition, less amounts of data and less bandwidth are required because the data volume required for localization is smaller than before due to the use of only one map layer instead of one map layer to date per sensor type.The method is advantageously used in certain areas, such as in the densely built urban environment, in which a satellite- and / or sensor-assisted position determination is not reliable enough.List of reference characters1 Vehicle 10 Environment sensor system 101 Detection range Environment sensor system 11 In-vehicle computing unit 2.1-2.3 Static features 3 External processing device (cloud)

Claims

Method for locating an at least partially automated driving vehicle (1) using its environment sensor system (10), wherein, in order to carry out the locating, the vehicle (1) is provided with an HD map with map information, at least comprising information on lanes, wherein the HD map contains information on geometric features (2.1-2.3) of static objects (2.1-2.3) in the vicinity of the lanes, which features can be detected by each sensor type of the environment sensor system (10), and wherein the locating is carried out in such a way that the vehicle continuously determines its current position by carrying out an environment detection by means of its environment sensor system (10) and matches geometric features (2.1-2.3) of detected static objects at the current position assumed by the vehicle with the geometric features (2.1-2.3) of the same static objects stored in the HD map, wherein, in the case in which the comparison reveals that a predefined degree of agreement is present for at least one geometric feature (2.1-2.3) recognized by the environment detection, the current position of the vehicle (1) is confirmed or determined as the actual vehicle position.The method according to claim 1, wherein the geometric features (2.1-2.3) of the static objects comprise at least: an edge of a building or a traffic sign, a corner of a building or a traffic sign, a window frame, a post of a sign or a traffic sign, a pavement marking, a curb edge, a barrier.Method according to Claim 1 or 2, wherein, in the event that the extent of agreement falls below a predefined limit value, a measure is taken.The method of claim 3, wherein the measure is one or a combination of: performing the localization again, calibrating a position sensor of the vehicle, matching with an alternative redundant localization, stopping the vehicle (1).Method according to one of the preceding claims, wherein a limit value for the predefined degree of agreement is defined in advance for predefined lanes.Method according to one of the preceding claims, wherein the geometric features (2.1-2.3) of the static objects detected by the environment sensor system are divided into geometric features (2.1-2.3) of the static objects close to the roadway and located above the roadway, processed and stored in separate HD maps.Method according to Claim 6, wherein weighting of the use of the HD maps takes place as a function of the current driving situation.Computer program which is configured to execute individual steps or all steps of the method according to one of Claims 1 to 7 by means of program code on a vehicle-internal computing unit (11).Computing unit (11) of a vehicle (1) on which the computer program according to claim 8 is implemented.Vehicle (1) driving at least in a semi-automated manner, having at least one environment sensor system (10) for detecting the environment and a vehicle-internal computing unit (11), on which the computer program according to Claim 8 is implemented.

Images