Method for estimating the measurement inaccuracy of an environmental sensing sensor

Abstract

Method for estimating measurement uncertainties of at least one environment sensing sensor (2) of an Ego vehicle comprising the following steps: - Recording (S1) of the environment of the Ego vehicle using at least one environment detection sensor (2); - Detecting (S2) at least one object located in an area in front of the ego vehicle in the direction of travel, - Defining (S3) a sensor output of at least one environment detection sensor (2) as ground truth when a definable minimum distance between the ego vehicle and the detected object is not reached, - Calculating (S4) a position of the object in relation to the ego-vehicle at an earlier time based on data from a positioning system (3); - Comparing (S5) a sensor output at the earlier time with the calculated position of the object; - Estimating (S6) the measurement uncertainty based on a result of the comparison.

Description

[0001] Many fusion systems use statistical estimators to fuse and / or track objects or raw data. One of the best-known methods is the Kalman filter and its derivatives. In many of these estimators (especially the Kalman filter), part of the fusion / tracking input is the measurement uncertainty of the sensor data.

[0002] There are essentially two methods that can be used to determine this measurement uncertainty a priori. First, the sensor manufacturer's measurement uncertainty information is used and entered according to the setup. These are then calculated, for example, according to ISO / IEC Guide 98-3. Second, the required measurement uncertainty is compared through trial and error or with ground-truth data and then statistically estimated.

[0003] A disadvantage of the aforementioned methods is that the information provided by sensor manufacturers refers to the measurement inaccuracy of the sensor itself, but not to the specific structure / use of the sensor in the system. Furthermore, especially in fusion systems with preprocessing (e.g., single-sensor object tracking, data stabilization, etc.), the measurement inaccuracies are altered by this preprocessing.

[0004] Furthermore, a position determination method in a vehicle navigation or mapping system is known from the prior art from the document US 2014 / 0 379 254 A1.

[0005] The document US 2009 / 0 228 204 A1 describes a system and a method for comparing a map with sensor-detected objects.

[0006] The document US 11 145 146 B2 shows a system for self-diagnosis of errors in an assisted or automated driving system of a vehicle.

[0007] WO 2018 / 031 678 A1 describes, among other things, a method for localization and mapping.

[0008] US 2020 / 0 309 541 A1 discloses a system for training a SLAM model, comprising a camera mounted in a vehicle and in communication with an image server via a cellular connection.

[0009] It is therefore an object of the present invention to provide a method which enables a reliable and accurate determination of the measurement inaccuracy of an environment detection sensor.

[0010] This object is preferably achieved by the subject matter of independent claim 1 and independent claim 8. Further advantageous embodiments of the invention are the subject matter of the dependent claims.

[0011] Initial considerations were that, for example, the position inaccuracies of extracted pedestrians from a camera image depend not only on the measurement inaccuracy of the camera itself, but also on the extraction algorithm, the assumptions used, the installation, position, etc. of the camera, and the lens used. As another example, radar objects are used in the environment model that have already been pre-tracked by a Kalman filter, meaning that the pure sensor measurement inaccuracies are no longer directly applicable to downstream fusion.

[0012] In all cases, the problem is that the exact measurement inaccuracies of the sensor outputs are unknown, which degrades the fusion results. Likewise, under these conditions, it is not possible to use the measurement inaccuracies specified by the manufacturer of the respective sensor. However, an exact determination of the measurement inaccuracies is complex and, to date, technically impossible. Therefore, a fundamental idea of ​​the invention was to use a statistical estimation to estimate the measurement uncertainty.

[0013] According to the invention, a method for estimating measurement uncertainties of at least one environment detection sensor of an ego vehicle is therefore proposed, comprising the following steps: - Recording an environment of the ego vehicle by means of at least one environment detection sensor; - Detecting at least one object located in an area in the direction of travel in front of the ego vehicle, - Setting a sensor output of at least one environment detection sensor as ground truth when a definable minimum distance between the ego vehicle and the detected object is undershot, - Calculating a position of the object relative to the ego vehicle at an earlier time based on data from a positioning system; - Comparing the sensor output at the earlier time with the calculated position of the object; - Estimate the measurement uncertainty based on a comparison result.

[0014] The environment detection sensor can be a radar sensor, a camera, or a lidar sensor, for example. It is also conceivable to use multiple sensors, whereby the sensors can be different or identical. For example, a combination of radar and camera could be used, whereby any combination of identical and different sensors is also conceivable. The method is carried out accordingly for the sensors. It is also conceivable to use multiple sensor outputs from one sensor instead of one sensor output, to calculate an average value and define this as the ground truth. Even when using multiple sensors, an average of the sensor outputs of the individual sensors can be used as the sensor output, for example, if a minimum distance to the detected object has been reached for all sensors.

[0015] This is based on the assumption that extraction or positioning is more accurate close to the vehicle, i.e., at a shorter distance, than at long distances. Therefore, using such a minimum distance is particularly advantageous.

[0016] The area in front of the ego vehicle includes not only the area directly in front of the vehicle but also areas to the sides of the vehicle. The object is determined by recording the environment detection sensor.

[0017] The sensor output at the earlier time includes a detection of the object and thus the detected position at that time. This sensor output is then compared with the calculated position.

[0018] Another basic assumption is that the position determination or calculation is more accurate and subject to fewer errors than the sensor output.

[0019] This allows for sensor- and setup-specific estimates of the sensor measurement inaccuracy to be made, which can later be used, for example, in a fusion system.

[0020] In a further preferred embodiment, a real-time kinematics system or odometry system is used for position determination. The odometry system uses the number of wheel revolutions to determine the position and orientation of the ego vehicle. The RTK (real-time kinematics) system uses GNSS satellite signals from geodetic receivers for position determination. The transmitters / receivers or sensors required for the respective position determination are provided in the ego vehicle according to the embodiment. For example, a wheel speed sensor is installed for the odometry system.

[0021] Furthermore, in a particularly preferred embodiment, the object is another road user, an environmental feature, or a landmark. It would also be conceivable to determine multiple road users and their distances. Furthermore, it would also be conceivable to detect a combination of the aforementioned objects and determine the distance to the ego vehicle in each case.

[0022] Particularly preferably, the estimated measurement uncertainty is provided to a sensor fusion system and / or a driver assistance system. This is particularly advantageous because the driver assistance systems and sensor fusion systems can factor in the measurement uncertainty, thereby increasing the accuracy and reliability of these systems.

[0023] In a further preferred embodiment, in addition to determining the distance, at least one angle between the object and the ego vehicle is determined. This is advantageous because it also allows for the consideration of objects, road users, or similar that are not directly in front of the ego vehicle. Furthermore, the angle between the object and the ego vehicle, or more precisely between the object and the direction of travel of the ego vehicle, influences the measurement inaccuracy. Accordingly, the measurement inaccuracy can also be greater at a larger angle.

[0024] In a further preferred embodiment, additional environmental factors are taken into account when determining the measurement uncertainty. Depending on the prevailing environmental factors, the sensor(s) used may be affected accordingly.

[0025] Furthermore, environmental factors such as current weather conditions or time of day are preferably considered. For example, measurement inaccuracy may be higher in rain than in clear weather. Rain, for example, can affect the detection of cameras and radar sensors, so it is advantageous to consider such conditions when estimating measurement inaccuracy. Likewise, the detection accuracy of a camera may be worse at night than during the day, or it may be affected by stray light in the evening due to the low position of the sun. Accordingly, it is advantageous to consider such environmental factors as well.

[0026] Furthermore, data on angles, weather conditions, and time of day can be stored and provided, for example, to a fusion system when a corresponding driving scenario, such as rain and night driving, is detected. This advantageously allows measurement inaccuracy to be provided right from the start.

[0027] The data obtained by the method is collected and processed in a computing device. Sensor data from a large number of journeys, under different weather conditions and times of day, is collected, and based on this data, a large number of measurement uncertainties are estimated for different scenarios. This large number of measurement uncertainties can then be provided to a vehicle so that, upon detection of a specific scenario, the vehicle can determine the correct measurement uncertainty without having to process a large amount of data within the vehicle itself.

[0028] The process will be illustrated using a radar sensor as an example. The radar sensor measures the position of other road users. For the process, a static road user is selected towards which the vehicle is moving. As soon as the vehicle falls below a minimum distance from the static road user, the sensor output is defined as ground truth. It would also be conceivable to calculate an average from several radar data or sensor outputs from the radar and define this average as ground truth. Using data from a positioning system, this position can now be calculated in relation to the ego vehicle at an earlier point in time and compared with the sensor measurement at that time. This technology can be applied to any number of road users and for any small / large distances as well as for any angles between the road users and the ego vehicle.Furthermore, environmental factors such as weather conditions and / or time of day can also be taken into account.

[0029] According to the invention, a system for determining measurement uncertainties is further provided, comprising at least one environment detection sensor, by means of which the environment of an ego vehicle is recorded and an object can be detected, a system for determining position, and a processor which receives a plurality of measurement uncertainties from an external computing device, and wherein the computing device is designed to determine a measurement uncertainty from the plurality of measurement uncertainties based on the sensor data. Furthermore, further factors in the vehicle or by the processor, such as weather conditions, time of day, and / or angle to the detected object, can be used to determine the measurement uncertainty. Based on this data, the corresponding measurement uncertainty is determined and selected from the plurality of measurement uncertainties. The processor and the external computing device are communicatively connected, for example, via a wireless data connection.

[0030] The processor can be located in a central control unit, such as the ECU. It would also be conceivable for the processor to be located in the environment detection sensor. If there are multiple environment detection sensors, the processor could be located in one of the sensors. The environment detection sensor, the processor, and the positioning system are connected to each other via a data link to enable data transmission.

[0031] Depending on how the estimated measurement uncertainty is further used, the processor can additionally be connected to a driver assistance system in order to forward the measurement uncertainty to the driver assistance system.

[0032] In a preferred embodiment, the position determination system is a real-time kinematics system or odometry system.

[0033] Further advantageous embodiments of the invention are the subject of the drawings, in which: Fig. 1: a schematic flow diagram of the method according to an embodiment of the invention; Fig. 2: a schematic representation of a system according to an embodiment of the invention.

[0034] In Fig. 1 shows a schematic flow diagram of the method according to an embodiment of the invention. In step S1, an environment of the ego vehicle is recorded using at least one environment detection sensor 2. In step S2, at least one object is detected that is located in an area in front of the ego vehicle in the direction of travel. In step S3, a sensor output of at least one environment detection sensor is defined as ground truth if a definable minimum distance between the ego vehicle and the detected object is undershot. In step S4, a position of the object in relation to the ego vehicle at an earlier time is calculated based on data from a position determination system 3. In step S5, a sensor output at the earlier time is compared with the calculated position of the object. Finally, in step S6, the measurement inaccuracy is estimated based on a result of the comparison.

[0035] Fig.2 shows a schematic representation of a system 1 according to an embodiment of the invention. The system 1 comprises at least one environment detection sensor 2, a position determination system 3, and a processor 4. The environment detection sensor 2 and the position determination system 3 are connected to the processor 4 via a data connection D. The data connection D can be wired or wireless. In this representation, the processor 4 is arranged, for example, in a central control unit, e.g., the ECU. It would also be conceivable for the processor 4 to be arranged in the environment detection sensor 2 for executing the corresponding method steps. List of reference symbols 1 system 2 environment detection sensor 3 Positioning system 4 processors D Data connection S1-S6 process steps

Claims

[1] Method for estimating measurement uncertainties of at least one environment detection sensor (2) of an ego vehicle, comprising the following steps: - recording (S1) an environment of the ego vehicle by means of at least one environment detection sensor (2); - detecting (S2) at least one object which is located in an area in the direction of travel in front of the ego vehicle, - setting (S3) a sensor output of at least one environment detection sensor (2) as ground truth when a definable minimum distance between the ego vehicle and the detected object is undershot, - calculating (S4) a position of the object in relation to the ego vehicle at an earlier time based on data from a position determination system (3); - comparing (S5) a sensor output at the earlier time with the calculated position of the object; - Estimate (S6) the measurement uncertainty based on a result of the comparison. [2] Method according to claim 1, characterized by that a real-time kinematics system or odometry system is used to determine the position. [3] Method according to one of the preceding claims, characterized by that the object is another road user, an environmental feature or a landmark. [4] Method according to one of the preceding claims, characterized by that the estimated measurement uncertainty is provided to a sensor fusion system and / or a driver assistance system. [5] Method according to claim 1, characterized by that in addition to determining the distance, at least an angle between the object and the ego vehicle is determined. [6] Method according to claim 1, characterized by that other environmental factors are taken into account when determining the measurement uncertainty. [7] Method according to claim 6, characterized bythat environmental factors such as current weather conditions or time of day are taken into account. [8] A system (1) for estimating measurement uncertainties comprising at least one environment detection sensor (2) by means of which an environment of an ego vehicle is recorded and an object is detectable, a position determination system (3) and a processor (4) which receives a plurality of measurement uncertainties from an external computing device, and wherein the computing device is designed to determine a measurement uncertainty from the plurality of measurement uncertainties based on sensor data from a plurality of journeys, wherein the processor (4) is further designed to define a sensor output of at least one environment detection sensor (2) as ground truth when a definable minimum distance between the ego vehicle and the detected object is undershot, to calculate a position of the object in relation to the ego vehicle at an earlier time based on data from the position determination system (3),to compare a sensor output at the earlier time with the calculated position of the object and to estimate a measurement inaccuracy based on a result of the comparison. [9] System according to claim 8, characterized by that the position determination system (3) is a real-time kinematics system or odometry system.

Images