Method and device for determining an integrity value for the position of a vehicle

Abstract

The invention relates to a method for determining an integrity value for the position of a vehicle and for an evaluation by means of an assistance system. According to the invention, the following steps are carried out: - determining semantic features relating to objects detected by means of a surroundings sensor system and detecting semantic features of objects at the position, determined by means of a localization method, from a map (S1); - mapping the semantic features onto defined object classes (S2); - linking the detected object classes of the surroundings to corresponding object classes from the map (S3); - assigning a specified frequency value pAH to each of the linkable object classes (S4); - adapting the integrity value of the vehicle position on the basis of the frequency value pAH (S5); - assigning a probability value for a false-positive detection of the object class for object classes that cannot be linked (S6); and - adapting the integrity value of the vehicle position on the basis of the probability value (S7). A deactivated assistance function is activated as soon as the integrity value exceeds a threshold value classified as being trustworthy (S8) or an activated assistance function is deactivated as soon as the integrity value falls below a threshold value classified as being trustworthy (S8).

Description

[0001] Method and device for determining an integrity value for a vehicle's own position

[0002] The invention relates to a method for determining an integrity value for a self-position and evaluation by a vehicle assistance system, as well as a device with a computing unit for determining an integrity value for a self-position and a vehicle assistance system configured for receiving and evaluating the determined integrity value.

[0003] Patent EP 3 130 945 B1 discloses a method for determining a vehicle's own position by identifying objects from a database at a position determined using a localization method, and by identifying objects in the vehicle's surroundings using environmental sensors. By comparing the positions of objects determined using different methods, the reliability of a position determined by GNSS can be assessed.

[0004] From the patent application DE 102018210 765 A1, a method is known in which landmarks are detected using recorded environmental data, whereby a self-position is determined based on the landmarks recorded in map data, which is used as a basis for determining a start and end point as well as a trajectory planning for an automated ferry operation.

[0005] High reliability, i.e., a corresponding integrity value of the determined self-position, is a prerequisite for safe automated ferry operation. However, since road sections look very similar and can contain many repetitive infrastructure elements and features, even a clearly incorrect position can appear plausible for a long time.

[0006] The object of the invention is therefore to provide a method and a device by which the reliability of a geolocation can be determined. This object is achieved by a method for determining an integrity value for a vehicle's own position and evaluating it by an assistance system with the features of claim 1 and an associated device according to claim 9. The dependent claims define preferred and advantageous embodiments of the present invention.

[0007] The inventive method involves the following steps: - Determination of semantic features of objects detected by means of an environmental sensor and

[0008] Capturing semantic features of objects at their own position determined by a localization method from a map, i.e. from a database comprising a map;

[0009] - Mapping of semantic features to defined object classes;

[0010] -Linking the object classes detected by means of environmental sensors with corresponding object classes from the map;

[0011] -Assignment of a predefined frequency value PAH to each of the linkable object classes;

[0012] -Adjustment of the integrity value of the vehicle position depending on the frequency value PAH;

[0013] -for non-linkable object classes, assignment of a probability value ppp for a false positive detection of the object class and

[0014] -Adjustment of the integrity value of the vehicle position depending on the probability value ppp;

[0015] -an integrity value above a threshold is considered trustworthy and where

[0016] -a disabled assistance function is activated as soon as the integrity value exceeds a threshold classified as trusted, or an activated assistance function is deactivated as soon as the integrity value falls below a threshold classified as trusted.

[0017] The accuracy of a self-position determined by a localization method known from the prior art is safety-relevant for an automated driving function and must therefore be ensured. The method according to the invention is implemented as a post-processing step following an upstream localization, which can be configured arbitrarily. Digital road maps are an integral component of many automated driving functions and, in addition to information about the topologies and attributes of the road segments, contain semantic features of objects recorded on the map. Similarly, semantic features of objects detected by environmental sensors such as cameras, radar, lidar, and / or ultrasound are determined. The semantic features of the objects detected and recognized by the environmental sensors are determined from databases, by means of image recognition, or by means of pre-trained machine learning models.Both on the map side and on the environment capture side, semantic features or groups thereof are mapped to predefined object classes.

[0018] The object classes of the environment are assigned, as far as possible, to corresponding object classes of the map that match in type; that is, object classes of the same type are associated, provided they can be encompassed or encircled by a predefined area such as a capture circle.

[0019] For this to work, the object classes must be projected onto a common coordinate system using the vehicle's position on the map, preferably the vehicle's coordinate system. Projecting the object classes onto a common coordinate system ensures a fast association process.

[0020] If the object classes cannot be encompassed by the predefined area, they are marked as unassignable, i.e., unassociable. For example, if an object class is detected by the environmental sensors, but no corresponding object class can be found on the map within a predefined area of ​​the user's position, then the object class detected by the environmental sensors is defined as unassociable.

[0021] Each of the linkable object classes is assigned a frequency value, PAH, i.e., a relative frequency of occurrence, which is read from a pre-filled memory. The detections of the object classes should be tracked to ensure that only exactly one association attempt is made for each object class in the environment, even if the object classes are detected multiple times. The association attempt of a detected object class can, for example, be performed by sequentially checking the object classes projected from the map onto the vehicle coordinate system to see if at least one of them is compatible and sufficiently close to the detected object class. Whether the proximity is sufficient is determined by a threshold value, which is preferably based on the desired accuracy of the localization. The dataing of these frequencies of occurrence, PAH, also called pAH rate, can be done manually based on expert assessment.Preferably, however, it is determined by offline processing of all available HD map data. For this purpose, all road segments are loaded sequentially from the map database, extracted object classes are counted, and finally, these counts are divided by the summed counts of all object classes in the processed road network to obtain a relative frequency of occurrence on individual road segments. It is preferable to use individual divisors for groups of object classes to calculate the relative frequency. For example, it is advantageous to divide the counts of object classes based on erratic landmarks by the summed counts of only this first subcategory, and to divide the counts of object classes based on line segments by the summed counts of only this second subcategory.

[0022] This takes into account that these two subcategories have significantly different spatial distribution statistics. The higher the probability of occurrence of the object classes on a given section of the route, the greater the risk of confusion and the lower the integrity. In other words, the lower the frequency value, the higher the integrity. Accordingly, the integrity is increased as the frequency value decreases. Optionally, the integrity can be reduced again as the frequency value increases.

[0023] The probability values ​​(ppp) for false positive detection are determined based on empirical data, estimates, or benchmarks. These values ​​indicate the likelihood of objects being incorrectly identified as such. Ideally, global average probability values ​​should not be applied; instead, values ​​under adverse conditions should be used. For example, false positive detections of line markings are extremely likely near tar joints or asphalt edges. A false positive detection of a speed limit sign can occur due to a speed limit sign painted on a truck. Furthermore, a base probability must be added to the false positive probability value (ppp), also known as the ppp rate, for all object classes. This base probability reflects the likelihood that a particular object class is incorrectly not recorded on the map (e.g., on the order of 5-15%).

[0024] The higher the probability value ppp for the detection of an object group, the more likely it is that this detection is not associatable, even if the vehicle's own position is correct. Accordingly, the integrity value of the vehicle position is reduced depending on the probability value ppp, with the reduction being smaller the higher the probability value ppp. When an object class is detected, it is checked whether the object class can be found on the map at the corresponding location, starting from the vehicle's own position (association attempt). If no corresponding object class can be found on the map and no association is possible, then the integrity value is reduced based on the probability value ppp, assuming the position is incorrect.An alternative assumption would be that the detection was incorrectly delivered and the object class cannot be found in the map, even if the self-position is correct. The higher the ppp probability value is assumed, the more likely this alternative assumption is for the failed association, and thus the probability of an incorrect self-position is lower. Therefore, the higher the ppp probability value, the less the integrity score is reduced.

[0025] The integrity value is initialized, for example, with a value expressing that the correctness / incorrectness of the position is equally probable. With each association attempt, the integrity value is updated so that the new integrity value expresses a joint probability derived from the previous integrity value I and an integrity value l. A, which can be derived solely from this association attempt. For example, in a representation as log-odds, i.e., as the logarithm of the probability ratio that the position is correct or incorrect, the updated integrity value in an association attempt is then l ne u = Lit, + IA. In the case of a successful association, IA is based on complete belief in the correctness of the position, minus the probability that the association arose by chance; for this, the occurrence probability PAH of the object class from the AH value table is used. In the case of an unsuccessful association, IA is based on complete belief in the incorrectness of the position, minus the probability that the non-associable detection arose by chance; for this, the false detection rate, i.e., the probability value ppp of the object class from an FP value table, is used.

[0026] In another embodiment, the threshold values ​​depend on the respective assistance function. Advantageously, an individual, reliable threshold value is defined so that each assistance function has optimal availability and is not deactivated more often than necessary due to unnecessarily high threshold values. In a further embodiment of the method, the threshold values ​​for activating and deactivating the assistance function have different values. This choice of threshold values ​​prevents the assistance function from being constantly activated or deactivated by toggling around the threshold value.

[0027] In a further refinement, semantic features of objects with variable parameters are subdivided into several object classes. An object class is a clearly defined instance and / or combination of detectable features, e.g., semantic features, which are mapped into several object classes defined by different value ranges. If a detected feature has variable parameters, it is advisable, for the sake of the most reliable integrity value possible, to define several object classes for different value ranges, for example: posts up to 1.2 m high, posts above 1.2 m high, line segment with curvature 0.0005 to 0.001, line segment with curvature 0.001 to 0.002, etc.

[0028] In a further development, the predefined frequency values ​​of the object classes are generated by evaluating maps and stored in a memory. The frequency values ​​can thus be continuously updated as the maps change. Preferably, the frequency values ​​are determined decentrally, for example on a server, and stored in the vehicle's memory via an wireless interface.

[0029] In another embodiment, the localization method determines the vehicle position using GNSS and vehicle sensors to determine environmental features and apply odometry. Determining the vehicle's own position using multiple sources improves starting conditions even before determining the integrity value according to the proposed method.

[0030] In a further development of the procedure, the integrity value is reset to its initial value if a localization inconsistency is detected. In the event of an inconsistency, the procedure is restarted, and corresponding assistance functions are deactivated to prevent errors or accidents. An inconsistency exists when there is doubt as to whether the latest position calculation is based on a continuation of the previous one. An inconsistency can be detected, for example, by an implausibly large jump or implausible fluctuation in the vehicle's position trajectory. The inconsistency is caused, for example, by missing or contradictory input data. Preferably, the vehicle localization system detects such conditions itself and outputs corresponding status signals.

[0031] The device according to the invention comprises a computing unit for determining an integrity value for a self-position and an assistance system of a vehicle configured for receiving, evaluating and applying the determined integrity value, wherein the computing unit is configured

[0032] - to determine semantic features of objects detected by means of environmental sensors and semantic features of objects at their own position determined from a map using a localization method;

[0033] - to map semantic features to defined object classes;

[0034] - to link the object classes detected by the environmental sensors with corresponding object classes from the map;

[0035] -assign a predefined frequency value PAH to each of the linkable object classes and adjust the integrity value of the vehicle position depending on the frequency value PAH;

[0036] -For non-linkable objects, assign a probability value ppp for a false positive detection of the object class and adjust the integrity value of the vehicle position depending on the probability value PFP; wherein the computing unit transmits the integrity value to an assistance system that activates a deactivated assistance function as soon as the integrity value exceeds a threshold classified as trustworthy or deactivates an activated assistance function as soon as the integrity value falls below a threshold classified as trustworthy.

[0037] Further advantages, features, and details will become apparent from the following description, in which—possibly with reference to the drawing—at least one embodiment is described in detail. Features described and / or illustrated can, individually or in any meaningful combination, constitute the subject matter of the invention, possibly also independently of the claims, and can, in particular, also be the subject of one or more separate applications. Identical, similar, and / or functionally equivalent parts are designated with the same reference numerals.

[0038] The device advantageously enables the integrity of a vehicle's own position, determined using a state-of-the-art method, without the need for additional components or data services.

[0039] This shows:

[0040] Fig. 1 Architecture for determining a self-position and

[0041] Fig. 2 Flowchart of the method according to the invention.

[0042] Figure 1 shows a prior art architecture for determining a vehicle's own position. The vehicle includes a satellite receiver 1 (GNSS), such as GPS, for determining its own position. This position, generally described by latitude and longitude, is located on a high-definition map (HD map) so that objects in the vicinity of the position and their semantic features can be determined from the map data 7. The relevant portion of the HD map data 7 for determining the position is downloaded from an external server 5 via an overhead interface.

[0043] Furthermore, the vehicle incorporates extensive sensor technology 3. An environmental sensor system 3a detects objects in the vehicle's vicinity; this environmental sensor system 3a is implemented, for example, as a camera, radar, lidar, and / or ultrasound. Using additional sensors 3b, such as wheel speed sensors and steering angle sensors, the vehicle's position and orientation are estimated based on odometry, starting from a point and using the detected wheel rotations and known kinematics.

[0044] The map data and the data acquired by sensors 3a and 3b are fed to a processing unit 9 for evaluation. Using the map data 7 and the data from sensor 3, the self-position determined via GNSS and, if applicable, other localization methods is checked, validated, and corrected.

[0045] Using the self-position in the HD map, information from the HD map and environmental sensors 3a is transferred into a common coordinate system 11 and used to control an assistance system 12, such as an automated driving system. For such assistance systems 12, it must be ensured that the self-position determined in the processing unit 9 is sufficiently reliable.

[0046] In order to be able to assess the reliability of the determined self-position, an integrity value is preferably determined in the computing unit 9 according to the procedure shown in Fig. 2 and a function of the assistance system 12 is activated or deactivated depending on the integrity value.

[0047] In a first step (S1), semantic features are determined for objects located in the vicinity of the determined self-position using the HD map data. These features are read from a database. The semantic features include precise descriptions of the objects, enabling largely error-free identification. Furthermore, semantic features are determined for objects detected by environmental sensors (3a). For this purpose, the sensor data is preferably evaluated using pre-trained models.

[0048] In a further step S2, the semantic features are mapped to defined object classes. The object classes can, for example, include the detected object itself, such as a 100 km / h speed limit sign, or a group of objects if this creates a characteristic element, for example, a double marking (dashed or solid), or a drivable hard shoulder (i.e., the group consisting of the last marking and the edge of the asphalt, if these are more than 2 m apart).

[0049] In the following step S3, the detected object classes of the environment are linked with corresponding object classes from the map; that is, the object classes of the environment are associated with compatible object classes on the map. Each of the associated object classes is then assigned a frequency value in S4.

[0050] In step S5, the integrity value is adjusted depending on the associated frequency value. In the following example, it is assumed that the integrity value is represented as log-odds, i.e., as the logarithm of the ratio of the probabilities that the position is correct or incorrect.

[0051] The starting point is an initial value I of the integrity value, which is initialized to I = log(1 / 1) = 0. This value is changed to l during a successful association process. ne u = kit + IA, where

[0052] IA = log ((1-PAH) / (PAH)), where PAH is read from a stored table as the frequency value for the given object class. Preferably, PAH, represented as log odds, lies within a value range ppp < 0.5.

[0053] In step S6, a probability value for a false-positive detection of an object class is assigned to a non-associable object class, for example, because the detected object class in the environment cannot be assigned a compatible object class from the environment within a given range. In step S7, if the association is unsuccessful, the integrity value of the vehicle position is adjusted depending on the probability value.

[0054] Ineu = lait + IA with IA = log ((PFP) / (1 - PFP)) , where the probability value ppp for the given object class is read from a table of values ​​that includes empirical values, estimates, and / or benchmarks for probability. Preferably, ppp, when represented as log odds, lies within a range ppp < 0.5.

[0055] In step S8, a deactivated assistance function is activated as soon as the current integrity value l ne u exceeds a threshold classified as trusted or an activated assistance function is deactivated as soon as the current integrity value l ne u falls below a threshold considered trustworthy.

Claims

Patent claims 1. Method for determining an integrity value for a vehicle's own position and evaluation by an assistance system (12) comprising the following steps: - Determination of semantic features of objects detected by means of environmental sensors (3a) and Acquisition of semantic features of objects at their own position determined from a map using a localization method (S1); - Mapping of semantic features to defined object classes (S2); - Linking the object classes detected by means of the environmental sensors (3a) with corresponding object classes from the map (S3); - Assignment of a predefined frequency value PAH to each of the linkable object classes (S4); - Adjustment of the integrity value of the vehicle position depending on the frequency value PAH (S5); - for non-linkable object classes, assignment of a probability value ppp for a false-positive detection of the object class (S6) and - Adjustment of the integrity value of the vehicle position depending on the probability value ppp (S7); Where - a disabled assistance function is activated as soon as the integrity value exceeds a threshold classified as trusted (S8) or an activated assistance function is deactivated as soon as the integrity value falls below a threshold classified as trusted (S8).

2. Method according to claim 1, characterized in that the object classes are projected into a common coordinate system.

3. Method according to claim 1 or 2, wherein the threshold values ​​depend on the respective assistance function.

4. Method according to one of the preceding claims, characterized in that the threshold values ​​for activation and deactivation of the assistance function have different values.

5. Method according to one of the preceding claims, characterized in that semantic features of objects with variable parameters are mapped into several object classes defined by different value ranges.

6. Method according to one of the preceding claims, characterized in that the predefined frequency values ​​PAH of the object classes are created and stored by evaluating maps.

7. Method according to one of the preceding claims, characterized in that the localization method determines the self-position using GNSS and vehicle sensors to determine environmental features.

8. Method according to one of the preceding claims, characterized in that the integrity value is reset to the initial value in the event of an inconsistency in the localization.

9. Device for carrying out the method of one of claims 1 to 8 comprising a computing unit (9) for determining an integrity value for a self-position and an assistance system of a vehicle configured for receiving and evaluating the determined integrity value, wherein the computing unit (9) is configured - to determine semantic features of objects detected by means of an environmental sensor (3a) and semantic features of objects at their own position determined from a map (7) using a localization method; - to map semantic features to defined object classes; - to link the object classes detected by means of the environmental sensors (3a) with corresponding object classes from the map; - to assign a predefined frequency value PAH to each of the linkable object classes and to adjust the integrity value of the vehicle position depending on the frequency value PAH; -for non-linkable objects, assign a probability value ppp for a false-positive detection of the object class and adjust the integrity value of the vehicle position depending on the probability value ppp; wherein the computing unit (9) transmits the integrity value to an assistance system (12), wherein the assistance system (12) - activates a disabled assistance function as soon as the integrity value exceeds a threshold classified as trusted, or deactivates an activated assistance function as soon as the integrity value falls below a threshold classified as trusted.

Images