DETECTING DRIVING-RELEVANT SITUATIONS AT A LARGER DISTANCE

Abstract

A method for detecting a relevant region in the surroundings of an ego vehicle, in which a situation exists which is relevant to the driving and/or safety of the ego vehicle, from measurement data of a sensor which observes at least a portion of the surroundings, the measurement data being discretized into pixels or voxels and/or are suitably represented in some other way, the existence of the relevant situation being dependent on the presence of at least one characteristic object in the surroundings, and the resolution of the pixels, voxels and/or the other representation being insufficient for directly detecting the characteristic object, the measurement data being analyzed for the presence of a grouping of objects which contains the characteristic object, the resolution of the pixels, voxels and/or the other representation being sufficient for detecting the grouping. A region in which the grouping is detected is classified as a relevant region.

Claims

1. A method for detecting at least one relevant region in surroundings of an ego vehicle, in which a situation exists which is relevant to the driving and/or safety of the ego vehicle from relevant measurement data of at least one sensor, which observes at least a portion of the surroundings, the method comprising: discretizing the measurement data onto pixels or voxels and/or representing the measurement data in some other way, wherein an existence of the situation is dependent on a presence of at least one characteristic object in the surroundings, and wherein a resolution of the pixels, and/or the voxels, and/or the other representation being insufficient for directly detecting the characteristic object; analyzing the measurement data to detect a presence of a grouping of objects which contains the characteristic object, wherein the resolution of the pixels, and/or the voxels and/or the other representation is sufficient for detecting the grouping; and classifying a region in which the grouping is detected as a relevant region.

2. The method as recited in claim 1, wherein another vehicle is selected as the characteristic object, and the grouping encompasses as a further object, in addition to the other vehicle, at least one further other vehicle and/or at least one road boundary and/or at least one lane marking and/or at least one additional object characteristic of the grouping.

3. The method as recited in claim 1, wherein a vehicle component which is separated from a vehicle or a vehicle cargo item which is separated from the vehicle is selected as the characteristic object, and the grouping encompasses, as the further object, in addition to the component or the cargo item, at least one road boundary and/or at least one lane marking.

4. The method as recited in claim 1, further comprising: evaluating, from the measurement data, a change in size of the grouping over time, and/or a change in density of the grouping over time, and/or a relative speed of the grouping relative to the ego vehicle, and/or a relative speed of the grouping relative to a road, and wherein an assessment as to whether a region containing the grouping is a relevant region additionally depends on the change in size over time, and/or the change in density, and/or the relative speed.

5. The method as recited in claim 1, wherein, in response to a region having been classified as a relevant region, a sensor and/or a device which supports acquisition of the measurement data by the sensor, is actuated with a control variable to change a physical parameter of the measurement data acquisition, and subsequently further measurement data relating to the relevant region are acquired.

6. The method as recited in claim 1, wherein, in response to a region having been classified as a relevant region, a further sensor is actuated and/or incorporated to acquire further measurement data from the relevant region.

7. The method as recited in claim 5, wherein such a change in the physical parameter of the measurement data acquisition is selected that, after the change, the resolution of the pixels, and/or the voxels and/or the other representation is sufficient to detect the characteristic object, and/or a further sensor is selected whose resolution of the pixels, and/or the voxels and/or the other representation is sufficient to detect the characteristic object.

8. The method as recited in claim 1, wherein, to check for the presence of the grouping, a subset of the measurement data relating to a region of the surroundings of the ego vehicle that is located at a predefined minimum distance from the ego vehicle is pre-selected.

9. The method as recited in claim 7, wherein, from a plurality of objects and/or groupings detected on the basis of the measurement data, a portion of the measurement data that is attributed to the detected objects and/or groupings is ascertained on the basis of a model of a contrast mechanism, and/or further physical imaging models, for the acquisition of the measurement data.

10. The method as recited in claim 1, wherein, in response to a region having been classified as a relevant region, a physical warning mechanism which is perceptible to a driver of the ego vehicle is actuated, and/or a drive system is actuated, and/or a steering system is actuated, and/or a braking system is actuated, for avoiding negative consequences for the ego vehicle, for the driver or for other road users, and/or for adapting a speed of the ego vehicle and/or a trajectory of the ego vehicle.

11. A classifier module for detecting objects in measurement data which have been obtained from surroundings of an ego vehicle, the classifier module configured to receive the measurement data as input and to deliver, as output, probabilities that the measurement data indicate a presence of one or multiple examples of one or multiple entities from a predefined set of sought entities, the sought entities including: at least one grouping of multiple vehicles, and/or at least one grouping of one or multiple vehicles together with at least one road boundary and/or with at least one lane marking, and/or at least one grouping of at least one component separated from a vehicle and/or a cargo item of the vehicle together with at least one road boundary and/or with at least one lane marking and/or with at least one additional object characteristic of the grouping.

12. The classifier module as recited in claim 11, wherein the classifier module is trained or is trainable using learning input data and respectively associated learning output data indicating the sought entities which are discernible in the learning input data.

13. The classifier module as recited in claim 12, wherein the classifier modules includes at least one artificial neural network.

14. A data set with learning input data and associated learning output data for a classifier module for detecting objects in measurement data which have been obtained from surroundings of an ego vehicle, the classifier module configured to receive the measurement data as input and to deliver, as output, probabilities that the measurement data indicate a presence of one or multiple examples of one or multiple entities from a predefined set of sought entities, the data set including (i) at least one grouping of multiple vehicles, and/or (ii) at least one grouping of one or multiple vehicles together with at least one road boundary and/or with at least one lane marking, and/or (iii) at least one grouping of at least one component separated from a vehicle and/or a cargo item of the vehicle together with at least one road boundary and/or with at least one lane marking and/or with at least one additional object characteristic of the grouping, as the sought entities which are represented in the learning input data and learning output data.

15. A non-transitory machine-readable storage medium on which is stored a computer program containing machine-readable instructions which, when executed on a computer and/or on a control unit and/or on a classifier module, prompt the computer and/or the control device to: (i) upgrade the classifier module to a classifier module as for detecting objects in measurement data which have been obtained from surroundings of an ego vehicle, the classifier module configured to receive the measurement data as input and to deliver, as output, probabilities that the measurement data indicate a presence of one or multiple examples of one or multiple entities from a predefined set of sought entities, the sought entities including (a) at least one grouping of multiple vehicles, and/or (b) at least one grouping of one or multiple vehicles together with at least one road boundary and/or with at least one lane marking, and/or (c) at least one grouping of at least one component separated from a vehicle and/or a cargo item of the vehicle together with at least one road boundary and/or with at least one lane marking and/or with at least one additional object characteristic of the grouping, and/or (ii) perform a method for detecting at least one relevant region in surroundings of an ego vehicle, in which a situation exists which is relevant to the driving and/or safety of the ego vehicle from relevant measurement data of at least one sensor, which observes at least a portion of the surroundings, the method comprising: discretizing the measurement data onto pixels or voxels and/or representing the measurement data in some other way, wherein an existence of the situation is dependent on a presence of at least one characteristic object in the surroundings, and wherein a resolution of the pixels, and/or the voxels, and/or the other representation being insufficient for directly detecting the characteristic object; analyzing the measurement data to detect a presence of a grouping of objects which contains the characteristic object, wherein the resolution of the pixels, and/or the voxels and/or the other representation is sufficient for detecting the grouping; and classifying a region in which the grouping is detected as a relevant region.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0062] FIG. 1 shows a schematic diagram, not true to scale, of a use case for method 100.

[0063] FIG. 2 shows an exemplary embodiment of method 100.

[0064] FIG. 3 shows a schematic diagram, not true to scale, of groupings containing further objects 32, 33 which are not other vehicles.

DETAILED DESCRIPTION OF EXAMPLE EMBODIMENTS

[0065] In FIG. 1, the ego vehicle 1 is moving from left to right along a road 4 designed as a three-lane expressway. Surroundings 11 of ego vehicle 1 are detected by a sensor 21. Sensor 21 delivers measurement data 21a to a classifier module 5, which searches the measurement data for a range of predefined entities 51a through 51d, such as other vehicles for example. In each case, the classifier module outputs a probability 52a-52d=P (51a through 51d) that the sought entity 51a through 51d has been detected in measurement data 21a. An appropriate response is then initiated. For example, a warning mechanism 15a, a drive system 14b, a steering system 14c and/or a braking system 14d may be actuated.

[0066] In the example shown in FIG. 1, the situation 13, where the traffic was jammed, exists in surroundings 11 of ego vehicle 1. The fact that this situation 13 is relevant to the driving and the safety of ego vehicle 1 is based on the other vehicle in the same lane as ego vehicle 1 and closest thereto in the direction of travel. This other vehicle is characteristic object 31, which nevertheless cannot be detected at a large distance due to the limited pixel resolution of sensor 21.

[0067] Classifier module 5 searches measurement data 21a specifically for a grouping 3 in which characteristic object 31 is combined with, in this example, further other vehicles as further objects 32 and 33. A region in surroundings 11 of ego vehicle 1 in which this grouping 3 is found is classified as region 12 in which situation 13 exists. Since grouping 3 is spatially much larger than constituents 31 through 33 thereof, it may be detected much earlier in measurement data 21a for the same pixel resolution of sensor 21.

[0068] FIG. 2 illustrates, by way of example, the sequence of method 100. From measurement data 21a delivered by sensor 21, optionally in step 105 a subset is selected which belongs to a region of interest at a relatively large distance from ego vehicle 1. The thinking behind this is that the relatively complex detection of groupings is unnecessary if the constituents of the groupings may also be directly detected individually.

[0069] In step 110, grouping 3 is detected in measurement data 21a. In step 120, the conclusion is drawn therefrom that the region containing grouping 3 is a region 12 which is relevant to the driving and/or safety of ego vehicle 1.

[0070] However, measurement data 21a may also be analyzed in step 130 for a change in size 34a or a change in density 34b of grouping 3, for a relative speed 34c of grouping 3 relative to ego vehicle 1, and/or for a relative speed 34d of grouping 3 relative to road 4; grouping 3 may have been detected previously in step 110 but may also be detected within step 130.

[0071] In step 140, based on the presence of grouping 3 in a region in vehicle surroundings 11, together with the additional information 34a through 34d, the conclusion is drawn that this region is a region 12 which is relevant to the driving and/or safety of ego vehicle 1. The work of step 120 is therefore also performed within step 140 albeit on the basis of expanded information.

[0072] Regardless of how exactly knowledge about the relevant region 12 has been obtained, various further measures are now possible.

[0073] In step 150, sensor 21 and/or a device (not shown in FIG. 2) which supports this sensor 21 may be actuated with a control variable 23. Then, in step 160, further measurement data 21b relating to the relevant region 12 may be acquired by sensor 21.

[0074] In step 170, a further sensor 22 may be actuated to acquire further measurement data 22a relating to the relevant region 12.

[0075] In step 180, portion 24 of the measurement data 21a, 21b, 22a that is attributed to the detected objects 31 through 33 or groupings 3 may be synthetized for the purpose of comparison with the original measurement data. In this way, the model of the contrast mechanism, and/or further physical imaging models of the camera, may be refined for the acquisition of measurement data 21a, 21b, 22a, which is in turn beneficial to the analysis in steps 120 and 130.

[0076] Finally, in step 190, warning mechanism 14a, drive system 14b, steering system 14c and/or braking system 14d may be actuated.

[0077] FIG. 3 schematically shows further possible groupings 3, with the aid of which a characteristic object 31 may be made detectable indirectly. Another vehicle as characteristic object 31 is grouped with a first lane marking as a further object 32. A lost tire as characteristic object 31a is grouped with a second lane marking as a further object 32a. A pedestrian as a characteristic object 31b is grouped with a road boundary as a further object 33.