Patent Application
20240096112
The present application claims the benefit and/or priority of German Patent Application No. 10 2022 209 713.0 filed on Sep. 15, 2022, the content of which is incorporated by reference herein.
The invention relates to a method for creating a lane model as well as a system for executing the method.
It is known from the prior art that, for assisted and highly automated driving functions such as highway pilots or fully automated parking without driver intervention, sensor systems are required which, in addition to recognizing other road users, also identify lanes robustly at large distances. On the basis of the lane recognition, a robust assignment of road users to lanes is necessary.
In addition to individual sensors, e.g., the installation of a front camera or a front radar, these are also installed jointly. In this case, a surroundings model including the relevant road users is formed separately nowadays, from the camera sensor and the radar sensor. The surroundings model is combined and the plausibility thereof is checked at object level. In particular, the lane recognition is based purely on camera data since the radar sensor is inherently unable to detect any lane markings which are currently used. There are moves to equip lane markings with radar reflectors so that radar sensors may detect these as well.
In addition to camera-based lane recognition, it is also possible to detect lane markings with the aid of a lidar sensor.
The disadvantage of the previously known sensors and methods is that, e.g., in the case of camera-based lane recognition, the long-range lane is only depicted in the camera model with a few pixels, which makes the task of detecting the course of the lane in particular at large distances very demanding and frequently prone to errors. Due to these problems, it is frequently difficult to assign lanes to road users at large distances. A flat world is, as a general rule, the basic assumption for camera-based lane recognition. This is frequently violated which, in turn, leads to erroneous results and the problems already described.
Furthermore, the disadvantage of the radar reflectors in the lane markings approach is that structural changes are necessary to the markings on the road for this approach to work, and this involves not inconsiderable expense. Furthermore, they also have to be maintained and regularly replaced depending on their design.
Lane detection with lidar is accordingly problematic since lidar sensors still represent a high cost factor compared with pure camera or radar systems. What also matters is that the range of lidars for lane recognition is inherently limited.
It is therefore an object of the present disclosure to provide a method and a system, by means of which a lane model is created, which has an improved range as well as improved accuracy.
This object is addressed by the independent claims 1 and 4. Further advantageous configurations and embodiments are the subject-matter of the subclaims.
Initial considerations were that an enormous amount of progress has been made in the field of radar sensor technology in the last few years. In particular, not only may newly developed radar sensors measure horizontally nowadays, but they also have the capacity to measure elevation, making possible an accuracy of elevation measurement of up to one-tenth of a degree. In particular, improving the elevation resolution for radar systems for assisted and automated driving is particularly advantageous. In particular, knowledge of a roadway curvature is advantageous since, in the case of camera-based lane recognition, the basic assumption of a flat world is taken as a starting point, which is however not appropriate in all cases.
According to the present disclosure, a method for creating a lane model by means of at least one first and at least one second environment detection sensor of an ego-vehicle is therefore proposed, having the following steps of:
The first and the second environment detection sensors may be different types of sensors. This is advantageous since various types of sensors offer different advantages during the detection of specific features in the surroundings. Thus, a camera may, for example, detect the lane markings better than a radar sensor.
In general, the term “roadway” includes multiple lanes, also lanes which have a traffic direction opposite the direction of travel of the ego-vehicle and, if necessary, are also physically delimited by a lane boundary. The term “lane” may include one or more lanes having the same traffic alignment, and in the present case corresponding to the direction of travel of the ego-vehicle.
Accordingly, it is conceivable that the at least one first environment detection sensor is a radar sensor and the at least one second environment detection sensor is a camera.
It would also be conceivable to combine more than two environment detection sensors.
When extracting the detections of the first environment detection sensor, the previously evaluated sensor data are used and analyzed in terms of the corresponding detections. The detections may be located directly on the roadway and, consequently, be assigned to the roadway or, due to their spatial proximity to the roadway, be accordingly assigned to the roadway. For example, roadway or lane boundaries are always arranged in a spatial proximity to the roadway.
The ground curvature is a vertical change in the ground plane. Accordingly, when estimating the ground plane, a linear deviation from the flat world assumption may be estimated. Furthermore, it may be estimated whether it involves a rise or a fall in the ground plane.
The estimation of the ground plane and/or ground curvature may be provided in an arithmetic unit in which the estimation is fused with the sensor data of the at least one second environment detection sensor. It would also be conceivable that the estimation is provided directly to the at least one second environment detection sensor, so that the sensor data generated by the second environment detection sensor already take account of this estimation and the detections are accordingly corrected in the second sensor.
As a result of the fusion of the estimation of the ground plane and/or ground curvature as well as the sensor data of the at least one second environment detection sensor, an advantageous lane model may be created since this process not only considers the course of the lane, but also a change in height. Accordingly, a more precise lane model is provided which, in turn, provides advantages during the object detection since, if the height differences of the lane are known, the distance from the object and also the size of the object can be established more accurately.
It would also be conceivable for the recording of the environment detection sensors to be provided to a neural network. The network may be trained such that it knows curved roadways and can therefore process the detections and estimates accordingly. Consequently, the network may then correctly output, e.g., the roadway markings, as real-world coordinates.
In a configuration, the estimation of the ground plane and/or ground curvature is carried out based on sensor data of a radar sensor, wherein an elevation measurement is carried out by the radar sensor for this purpose. It is advantageous to use a radar sensor since it is possible, thanks to a radar sensor having elevation resolution or an elevation measuring capacity, to detect a change in roadway curvature of less than 0.5 m at a distance of 200 m.
It is further understood that the extracted detections of the at least one first environment detection sensor include objects on the roadway and/or at the edge of the roadway. Accordingly, in one configuration, the radar detections of the objects on the roadway such as, for example, other road users or objects which can generally be detected with the radar sensor as well as objects at the edge of the roadway such as, for example, a guardrail or peripheral developments are, in this case, extracted. In particular, the detections at the edge of the roadway, which constitute a roadway or a lane boundary are advantageous since, for example, the width of the roadway and the rough course of the roadway may thus already be determined.
Furthermore, a system for creating a lane model in an ego-vehicle is proposed according to the present disclosure, including at least one first and at least one second environment detection sensor for recording the environment of the ego-vehicle, an evaluation unit for evaluating the sensor data of the at least one first and at least one second environment detection sensor as well as an arithmetic unit for determining a roadway, for extracting detections of the at least one first environment detection sensor, which can be assigned to the roadway, for estimating a ground plane and/or ground curvature based on the sensor data of the at least one first environment detection sensor, for providing the estimation as well as for creating the lane model by fusing the estimation of the ground plane and/or ground curvature and the sensor data of the at least one second environment detection sensor.
The arithmetic unit can, for example, be a central control unit such as an ECU or an ADCU. It would also be conceivable for the arithmetic unit to be part of one of the sensors and for the corresponding method steps to take place in one of the sensors.
In a configuration, the at least one first environment detection sensor is a radar sensor.
In a further embodiment, the radar sensor is configured such that an elevation measurement may be carried out.
In a configuration, the at least one second environment detection sensor is a camera sensor. A camera sensor is, accordingly, in particular advantageous since, e.g., lane markings may be detected simply and precisely with a camera.
Further advantageous configurations and embodiments are the subject-matter of the drawings, wherein:
| Number | Date | Country | Kind |
|---|---|---|---|
| 10 2022 209 713.0 | Sep 2022 | DE | national |
