Method for extracting features from measurement data collected through crowdsourcing

Abstract

The invention relates to a method for determining features for a digital map, in particular a digital HAD map, by means of a control device, wherein determined measurement data of a sensor of at least one vehicle are received or called up; the measurement data are superimposed on one another in accordance with their geographic coordinates; the superimposed measurement data are correlated with one another and optimized with respect to errors determined by the superposition; the measurement data are clustered and features are extracted from the clusters formed; the extracted features are provided for updating or creating a digital map. The invention also relates to a control device, a computer program and a machine-readable storage medium.

Description

Technical Field

[0001] The invention relates to a method for determining features for a digital map, in particular a digital HAD map, by means of a control device. The invention also relates to a control device, a computer program and a machine-readable storage medium. Background Art

[0002] Today, vehicles that can operate autonomously and automated driving functions rely on up-to-date and accurate maps. Such maps are usually implemented as digital HAD (Highly Automated Driving) maps and have multiple layers.

[0003] The map may have, for example, a planning layer and a positioning layer. The planning layer is used to implement trajectory planning and has street directions and street geometry.

[0004] By using the localization layer, the vehicle can compare the features in its environment detected by sensors with the virtual features in the localization layer and determine its position in the planning layer. These features can be detected by radar sensors, for example.

[0005] The following methods are known so far: HAD maps with a positioning layer and a planning layer are created from measurement data from special measurement vehicles. In order to create and store the features of the positioning layer, measurement data from sensors of different types (such as radar sensors and camera sensors) are usually compared. Such measurement drives can only meet the requirements of HAD maps in terms of real-time performance and accuracy at a high cost. Summary of the invention

[0006] The object of the present invention can be seen as to be to specify a method for ascertaining or extracting features from measurement data which can be implemented with low technical complexity.

[0007] The object is achieved by means of the technical solution described below according to the invention.

[0008] According to one aspect of the present invention, a method for obtaining features for a digital map, in particular a digital HAD map, by means of a control device is provided. In one step, the obtained measurement data of at least one sensor of at least one vehicle is received or called. Preferably, it is conceivable to use measurement data of multiple vehicles, which are obtained by sensors of similar categories. The at least one sensor may be a radar sensor, a LIDAR sensor, an ultrasonic sensor, a camera sensor, etc. In addition, a combination of different sensors or the same sensor may be used.

[0009] Alternatively or additionally, only measurement data from a plurality of sensors of a vehicle may also be used for the method.

[0010] Depending on the configuration of the control device, the measurement data may already be present in a memory and be called up, or may be retrieved from one or more vehicles. The vehicle is understood in particular to be a mobile unit, which may be configured as a robot, a water vehicle, an air vehicle, etc. In particular, the term vehicle is not limited to motor vehicles.

[0011] In a further step, the measurement data determined by different vehicles are superimposed on one another according to their geographical coordinates. This step can be performed by a rough superposition of the measurement data with the aid of GPS information.

[0012] The GPS information can preferably include the vehicle position and the vehicle orientation. The alignment of the measurement data can be inaccurate due to noise and systematic errors. The superimposed measurement data are thus linked to one another and optimized with respect to errors. The optimization can eliminate errors determined by the superposition and minimize spatial deviations of the individual measurement data sets from one another.

[0013] In a further step, the measurement data of different measurement data sets are combined into object clusters and features are extracted from the individual object clusters. The measurement data sets or compressed measurement data clouds are detected by clustering and groups or clusters are formed. Each cluster can then be extracted in the form of features.

[0014] Next, the extracted features may be used or provided when creating or updating a planning layer and / or a positioning layer of a HAD map.

[0015] By optimizing and clustering the measurement data sets, feature extraction can be made more efficient and comparison with measurement data of other sensor types (eg camera sensors) can be omitted. In particular, the method makes it possible to determine and extract features by using only one sensor type.

[0016] Feature extraction can simplify object clusters or measurement data clusters (which map non-overlapping multiple objects detected by different vehicles) into a virtual feature or a virtual object for use in the HAD map. This step can be achieved, for example, by a clustering algorithm. The feature can be in particular an object. For example, the feature can be a traffic light, a sign, a columnar structure, a tree, etc.

[0017] The optimization of the measurement data set can be achieved, for example, by associating the measurement data with adjacent virtual objects in the planning layer. For the association, the properties or features of the planning layer can be called up in parallel with the measurement data. This step allows the received measurement data to be synchronized with the existing features of the planning layer.

[0018] According to another aspect of the present invention, a control device is provided, wherein the control device is configured to implement the method. The control device may be, for example, a control device external to the vehicle or a server unit external to the vehicle, such as a cloud system. The control device may preferably receive measurement data of at least one sensor and / or measurement data of multiple sensors.

[0019] In addition, according to one aspect of the present invention, a computer program is provided, the computer program comprising instructions, which, when executed by a computer or a control device, cause the computer program to perform the method according to the present invention. According to another aspect of the present invention, a machine-readable storage medium is provided, on which a computer program according to the present invention is stored.

[0020] The at least one vehicle can be operated in an assisted, partially automated, highly automated and / or fully automated manner or without a driver according to the BASt (Bundesanstalt für Straβenwesen: German Federal Institute for Transport Research) standard. In particular, the vehicle can be a mobile unit, which can be configured as a vehicle, a robot, a drone, a water vehicle, a rail vehicle, a robot taxi, an industrial robot, a commercial vehicle, a bus, an airplane, a helicopter, etc.

[0021] The method and control device according to the invention can simplify and speed up complex processing steps for extracting features for map creation or map updating. As a result, the measurement data collected by individual vehicles can be optimally utilized and used for planning the layers.

[0022] According to one embodiment, the measurement data or measurement data sets determined by the sensors of at least two vehicles are linked to at least one sensor-specific positioning map and geographically aligned. This approach allows a particularly precise superposition of the individual measurement data sets and thus a particularly precise feature extraction of properties or features for the planning layer.

[0023] According to another embodiment, the measurement data acquired by the sensors of at least two vehicles are geographically aligned by means of predetermined alignment attributes. For this purpose, special alignment attributes for the targeted alignment of the measurement data sets can be provided and used for the alignment and association of the measurement data sets. In addition to the features of the planning layer, the alignment attributes can be provided in addition to achieve alignment or correction of the measurement data to the existing features of the planning layer. This allows the data of the planning layer of the collection map to be separated from the alignment step of the individual data.

[0024] According to another embodiment, the measured data obtained from sensors of at least two vehicles and / or a fleet of vehicles are received or called by a control device external to the vehicle and the measured data are processed to extract features. In this case, the control device external to the vehicle can be configured as a server unit or a cloud. The vehicle can transmit the measured data or measurement data sets obtained to the control device continuously or at defined time intervals. For this purpose, a communication link based on WLAN, GSM, UMTS, LTE, 5G or similar transmission standards can be set between the vehicle and the control device. Crowdsourcing of the measured data is achieved by these measures, wherein the control device acts as a central unit.

[0025] According to another embodiment, the acquired measurement data are received by a vehicle-side control device and processed to extract features. As a result, the processing of the measurement data can be carried out at least partially in the individual vehicles from which the measurement data are collected. Here, the results of the feature extraction can also be sent to a central server unit or a cloud and stored there. The optimization and extraction steps can be carried out in particular outside the vehicle in order to use a higher stable computing performance. These measures can also centrally achieve the matching of algorithms for the association and optimization of the measurement data.

[0026] According to another embodiment, features in the form of objects are extracted and transferred to the planning layer of the map. The features can be configured in particular as traffic lights, signs, columnar structures, trees, etc. Preferably, objects are identified and extracted from clusters of measurement data, thereby also enabling localization across sensor categories.

[0027] According to another embodiment, the measurement data are determined by a radar sensor, a LIDAR sensor, an acoustic sensor or a camera sensor. Thus, the method is not limited to a specific sensor type. In particular, features can be extracted from the measurement data of different sensor types separately or in combination with multiple sensor types. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The preferred embodiment of the present invention is described in detail below based on a highly simplified schematic diagram. It is shown here:

[0029] Figure 1 : A flow chart illustrating a method according to one embodiment by comparison with a mapping drive;

[0030] Figure 2-4 : used to illustrate the steps of a method according to an embodiment;

[0031] Figure 5 : A flow chart for illustrating a method according to another embodiment;

[0032] Figure 6 : A flowchart illustrating a method with alignment properties used according to one embodiment. DETAILED DESCRIPTION

[0033] exist Figure 1 , a flow chart of a method 1 according to one embodiment is shown by comparison with a conventional mapping drive 2 .

[0034] In a conventional mapping drive 2, measurement data 6 are collected by a special measuring vehicle 4 and subsequently processed in order to extract features 8 for the HAD map. In addition to the method 1 according to the invention, the conventional mapping drive 2 can also be carried out.

[0035] In the method 1, measurement data or measurement data sets 10, 11 are collected from a plurality of vehicles 12, 13. The measurement data 10, 11 can be preprocessed 14, 15, for example, within the vehicle. The measurement data 10, 11 can be collected, for example, by radar sensors and / or LIDAR sensors on the vehicle side.

[0036] In a further step 16, the preprocessed measurement data 14, 15 are superimposed and correlated with one another. Due to noise and systematic errors, the alignment or superposition of the measurement data 10, 11 may be inaccurate. Therefore, the measurement data 10, 11 determined by the individual vehicles 12, 13 deviate from one another.

[0037] In this case, the preprocessed measurement data 14, 15 are transmitted to a control device 3 outside the vehicle and further processed. For this purpose, a communication link 7 can be provided between the vehicles 12, 13 and the control device 3, which is based on, for example, WLAN, GSM, UMTS, LTE, 5G and similar transmission standards. This measure allows crowdsourcing of the measurement data 10, 11, 14, 15, wherein the control device 3 acts as a central unit. The control device 3 is designed, for example, as a cloud, which can collect the measurement data 10, 11, 14, 15 by crowdsourcing and use them to extract features 20.

[0038] The extracted features 20 may be stored in a machine-readable storage medium 5 , for example.

[0039] In a further step 18 , the superimposed measurement data 16 are optimized. In this case, for example, errors and offsets can be eliminated and thus the deviation A of the superimposed measurement data 16 can be reduced. Subsequently, one or more features 20 can be extracted from the corrected measurement data 18 .

[0040] Figure 2 , Figure 3 and Figure 4The steps for explaining the method 1 according to one specific embodiment are shown by way of example. The determined or extracted feature 20 can be designed, for example, as a traffic light system. The traffic light system is detected by radar sensors (not shown) of vehicles 12 , 13 and evaluated by control device 3 in the form of measurement data 10 , 11 .

[0041] exist Figure 2 , superimposed measurement data 16 of different vehicles 12 , 13 are simultaneously shown in order to indicate a deviation A of the measurement data 16 determined by different vehicles 12 , 13 . Due to the lack of alignment, multiple traffic light installations may be extracted as features when creating a planning map.

[0042] Figure 3 A further step of the method 1 is shown, in which the superimposed measurement data 16 are corrected and optimized. This step reduces the deviation A of the respective measurement data 16 and compresses the traffic light positions ascertained by the respective vehicles 12, 13. In this case, a traffic light which is actually only one is still represented by a plurality of traffic lights ascertained by the sensors.

[0043] Figure 4 In a further step, features 20 are extracted from the compressed clusters of the measurement data representing the traffic lights. In this step, features 20 of the measurement data set are formed by means of a clustering algorithm and are thus reduced to unique objects.

[0044] For example, a feature 20 implemented as a traffic light can be used in the subsequent mapping of the planning layer and / or the positioning layer of the HAD map. In addition, multiple features can also be extracted simultaneously or successively by the method. For this purpose, the measurement data sets are identified, clustered and extracted as features 20. In principle, all objects can be used as features 20. Trees, columnar structures (such as street lights or gantries), traffic lights, signs, etc. can be used as features 20 in particular.

[0045] Figure 5 A flow chart is shown for illustrating a method 1 according to another specific embodiment. Measurement data 22 collected by crowdsourcing are received by a control device 3. For this purpose, measurement data sets of a plurality of vehicles are received. These vehicles can collect measurement data sets during regular driving and transmit them to the control device 3 or the cloud.

[0046] The measurement data are then aligned 24 according to the localization map or the localization layer L of the HAD map. In particular, measurement data sets determined from different vehicles can be aligned with respect to one another and superimposed.

[0047] The alignment 24 of the measurement data based on the localization layer L can then be used for clustering and extraction 26 of features 20 for the planning layer P.

[0048] Figure 6 A flow chart is shown for illustrating a method 1 according to another implementation method. The flow chart shows the use of alignment properties.

[0049] In one step, measurement data 22 collected by crowdsourcing, such as radar measurement data from different vehicles or a fleet of vehicles, are received. In parallel therewith, data 30 of a planning layer P with alignment properties 32 may be received.

[0050] The data of the planning layer P and the alignment properties 32 can also be determined by the vehicle through measurements or provided, for example, by a map manufacturer in order to facilitate frequent and automated updating of features 20 in the map.

[0051] Based on the collected measurement data 22 and the alignment properties 30, the collected measurement data 22 are aligned according to the positioning layer L. Next, the aligned measurement data are optimized and clustered. Next, features 20 may be extracted 26 from the measurement data.

Claims

1. A method (1) for ascertaining features (20) for a digital map by means of a control device (3), wherein: Receiving or retrieving ascertained measurement data (10, 11, 22) of at least one sensor of at least one vehicle (12, 13); superimposing the measurement data (10, 11, 22) on each other in accordance with their geographical coordinates; Correlating the superimposed measurement data (16) with one another and optimizing the superimposed measurement data with respect to errors determined by the superposition, wherein spatial deviations of the measurement data from one another are minimized by the optimization; Clustering the measurement data (16) and extracting features (20) from the formed clusters (18) to simplify the formed clusters into a virtual feature; providing the extracted features (20) for updating or creating a digital map, In which, before the superimposed measurement data (16) are optimized, the measurement data (10, 11, 22) obtained by sensors of at least two vehicles (12, 13) and collected by crowdsourcing are associated with at least one sensor-specific positioning layer (L) and geo-aligned, and / or the measurement data (10, 11, 22) obtained by sensors of at least two vehicles (12, 13) and collected by crowdsourcing are geo-aligned by predetermined alignment attributes (32), wherein the alignment attributes are based on data of a planning layer (P) and enable the measurement data to be aligned to existing features of the planning layer (P).

2. The method according to claim 1, wherein: Measurement data (10, 11, 22) acquired from sensors of at least two vehicles (12, 13) and / or a fleet of vehicles are received or retrieved by a vehicle-external control unit (3) and processed to extract features (20).

3. The method according to any one of claims 1 to 2, wherein: The ascertained measurement data (10, 11, 22) are received by a vehicle-side control unit (3) and processed to extract features (20).

4. The method according to any one of claims 1 to 3, wherein: Features (20) in the form of objects are extracted and transferred to a planning layer (P) of the map.

5. The method according to any one of claims 1 to 4, wherein: The measurement data (10, 11, 22) are ascertained by means of a radar sensor, a LIDAR sensor, an acoustic sensor or a camera sensor.

6. The method according to claim 1, wherein: The digital map is a digital HAD map.

7. A control device (3), wherein: The control device (3) comprises a memory, a processor and a computer program stored in the memory, wherein the computer program implements the method according to any one of claims 1 to 6 when executed by the processor.

8. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1 to 6.

9. A machine-readable storage medium (5) on which a computer program is stored, which, when executed by a processor, implements the method according to any one of claims 1 to 6.

Images