DE102015205087A1 - Method for determining the misalignment of a driver assistance sensor - Google Patents

Abstract

The invention discloses a method for determining a misalignment of a driver assistance sensor of a motor vehicle with the following steps: recording a first cloud of points with the driver assistance sensor, the first cloud of points imaging at least one stationary object in the surroundings of the motor vehicle; If the motor vehicle has moved on a predetermined distance or if a predetermined period of time has elapsed, recording a second cloud of points with the driver assistance sensor, the second cloud of points imaging the at least one stationary object in the vicinity of the motor vehicle; - transformation of the two point clouds into a common coordinate system; - stepwise turning the first cloud of points and the second cloud of points in the same direction until an abort criterion is met; and determining, after each incremental rotation of the first point cloud and the second point cloud, a measure of coincidence of the first point cloud and the second point cloud, the abort criterion being satisfied when the measure of the match has reached an optimal value; and determining a rotation of the driver assistance sensor about its z-axis on the basis of the rotation angle of the first cloud of points and the second cloud of points at which the degree of agreement has reached an optimum value.

Description

The present invention relates to a method for determining the misalignment of a driver assistance sensor, in particular about the z-axis during normal operation of a motor vehicle.

Today's driver assistance systems in a motor vehicle are based on the data of a plurality of sensors. These include radar sensors (Radio Detection and Ranging), LIDAR sensors (Light Detection and Ranging), camera sensors and ultrasonic sensors. These are used for example in a distance-regulating cruise control, a lane departure warning assistant and / or a parking assistant. In order to meet the requirements of future driver assistance systems, especially in highly automated driving, a restriction to a single sensor principle or measuring principle is not sufficient. In particular for highly automated driving, a redundancy on the basis of redundant sensor principles is absolutely necessary due to the required low error rate. In the field of vehicle environment detection laser pulses are increasingly used in the invisible infrared spectrum. When the laser pulses emitted by the sensor strike an object, part of the emitted laser radiation is reflected back to a receiving unit. By evaluating the duration of the laser pulse, the distance to the detected object can be determined by means of the speed of light. Such LIDAR sensors offer in comparison to other distance-measuring sensors, such as radar sensors, a higher angular resolution and are therefore particularly suitable for detecting object contours and the passable area or free space.

In a motor vehicle, currently mainly 2D LIDAR sensors and only to a lesser extent 3D sensors are used. Such a LIDAR sensor generates a point cloud in the angle information and distance information to which the detected reflection is stored.

For the efficient use of the data of a LIDAR sensor, the correct calibration of the LIDAR sensor is absolutely necessary. In addition to an intrinsic calibration, which includes the correction of internal sensor parameters and which is performed in the prior art by the manufacturer of the LIDAR sensor, an extrinsic calibration of the LIDAR sensor is required. In the case of the latter, the exact installation position and the installation position of the sensor in the motor vehicle are determined. The position and location of the LIDAR sensor may change during operation of the motor vehicle. Such a change is called a misalignment and describes a deviation from the original or predetermined installation position and installation position of the LIDAR sensor.

In the prior art it is known to calibrate the LIDAR sensor by means of a stereo camera and a known object. Furthermore, in the prior art, the pairwise calibration of LIDAR sensors is known.

The prior art methods require a known environment or object. Furthermore, they can not detect a misalignment of a LIDAR sensor, if only a single LIDAR sensor is present. When using a reference frame based on a stereo camera to calibrate the LIDAR sensor, it is disadvantageous that the LIDAR sensor uses sensor hardware other than the stereo camera. A disadvantage of this method is that a second sensor and a second evaluation device are required. Furthermore, the aforementioned test object is required. The pairwise calibration of two LIDAR sensors means that the LIDAR sensors can only be calibrated relative to one another but not relative to a vehicle coordinate system.

The DE 10 2004 056 400 A describes determining angular misalignment of a proximity sensor using previously known markers.

It is an object of the invention to provide a method for determining the misalignment of a driver assistance sensor of a motor vehicle during normal operation.

The object of the invention is achieved by a method according to claim 1. The dependent claims claim preferred embodiments.

The method for determining a misalignment of a driver assistance sensor of a motor vehicle comprises the step of recording a first cloud of points with the driver assistance sensor, wherein the cloud of points images at least one stationary object in the surroundings of the motor vehicle. If the motor vehicle has moved on for a predetermined distance or if a predetermined period of time has elapsed, a second cloud of points is recorded by the driver assistance sensor, the second cloud of points imaging the at least one stationary object in the surroundings of the motor vehicle. Both point clouds are transformed on the basis of the previously known installation position and installation orientation and the well-known movement of the vehicle between the two shots in a common coordinate system. The first cloud of points and the second cloud of points are shifted so that at least a stationary object is arranged at a non-de-adjusted driver assistance sensor in the common coordinate system at the same location. Subsequently, the first cloud of points and the second cloud of points around the respective sensor origin in the same direction are rotated stepwise until a termination criterion is met. After each stepwise rotation of the first cloud of points and the second cloud of points, a measure of the correspondence of the first cloud of points and the second cloud of points is determined, wherein the abort criterion is met when the degree of correspondence has reached an optimal value. Based on the angle of rotation of the first cloud of points and the second cloud of points at which the degree of agreement has reached an optimum value, a rotation of the driver assistance sensor about its z-axis is determined.

The method according to the invention makes it possible to check a horizontally aligned three-dimensional LIDAR sensor without a further sensor and without a calibration object in the surroundings of a motor vehicle moving in use with regard to a misalignment. The misalignment is determined in particular around the z-axis. The inventors have assumed that the installation position of the driver assistance sensor is known with sufficient accuracy from the design data of the vehicle. The method according to the invention uses the movement data of the motor vehicle as well as two unprocessed point clouds that were recorded with the driver assistance sensor. The method according to the invention has the advantage that it can be used without further knowledge about a model, for example via a calibration object, in the natural environment of a moving motor vehicle. This makes it possible to detect a misalignment of a driver assistance sensor during operation of the motor vehicle. The step of determining the match may include the step of minimizing a cost function. The step of minimizing the cost function may include the step of minimizing the mean square distance.

If the driver assistance sensor is not misadjusted, the first cloud of points and the second cloud of points coincide in an overlapping partial area. If the driver assistance sensor is misaligned about the z-axis, the two point clouds offset due to the movement of the vehicle. After turning the point cloud in the same direction, this offset disappears if both the first point cloud and the second point cloud have been rotated by the angle corresponding to the misadjustment angle of the driver assistance sensor.

The method may further comprise the step of transforming the first cloud of points and the second cloud of points into the world coordinate system prior to stepping the first and second cloud of points. In this case, the first cloud of points and / or the second cloud of points must be translated translationally in the direction of travel relative to the other. The predetermined distance, after which the second cloud of points is recorded, is about 3 m to about 10 m, preferably about 4 to about 6 m. The first and second cloud of points are rotated about the z-axis of the driver assistance sensor, preferably around the z-axis of the driver assistance sensor in the world coordinate system.

The first point cloud may have a different center of rotation than the second point cloud when rotated. Preferably, the center of rotation is the location of the driver assistance sensor in the world coordinate system when detecting the first cloud of points or the second cloud of points.

The method may also include the step of eliminating reflections from dynamic objects.

The driver assistance sensor may be a LIDAR sensor, preferably a two-dimensional LIDAR sensor, which is designed to detect the surroundings of the motor vehicle.

The driver assistance sensor may be directed in the direction of travel of the motor vehicle and detect at least one object in front of the motor vehicle. The driver assistance sensor may be directed transversely to the direction of travel and detect at least one object next to the motor vehicle. Alternatively or additionally, the driver assistance sensor may be directed opposite to the direction of travel of the motor vehicle and detect an object behind the motor vehicle. The driver assistance sensor may be arranged to detect objects in front of and in addition to the motor vehicle, or be arranged to detect objects behind and beside the motor vehicle.

The at least one object may be any stationary object in the vicinity of a roadway that passes the motor vehicle during a ride with a driver from a starting point to a destination point, wherein the object is not configured for adjusting the driver assistance sensor. The at least one object is preferably located in the natural environment that passes through a motor vehicle when driving outside of a workshop or maintenance room as intended. The stationary object can be, for example, a building, a roadway boundary, a protective barrier, also called guardrail, a guide post and / or a plant, for example a tree, be. The method may use any other stationary objects in the vicinity of the roadway.

The steps of detecting the first cloud of points and detecting the second cloud of points and determining the rotation about the z-axis may be performed repeatedly. This makes it possible to continuously monitor the misalignment of the driver assistance sensor. The successively detected twists about the z-axis of the driver assistance sensor can be filtered or averaged. This makes it possible to reduce measurement errors. The filtering can be done for example by means of a so-called median filter.

The method may further comprise the step of compensating for the rotation of the driver assistance sensor about the z-axis. Here, the installation position of the sensor during operation or when the vehicle is stopped by the specific misalignment angle is corrected. This can be done by an adjustment of the sensor or in software.

The invention also relates to a computer program product which, when loaded into a computer having a processor and a memory, performs the method described above.

The invention will now be elucidated with reference to the attached figures by means of a non-limiting and exemplary embodiment, wherein

1 shows a moving motor vehicle picking up two point clouds;

2 shows two point clouds in the world coordinate system rotated clockwise;

3 shows two point clouds in the world coordinate system, which were turned counterclockwise;

4 shows two point clouds in the world coordinate system, which are essentially on top of each other.

1 shows a motor vehicle 2 that moves in the direction of the arrow. At time t1, the motor vehicle is at location s1 and at time t2 the motor vehicle is at location s2. The location s1 and the location s2 have a distance of about 1 m to about 10 m, preferably from about 4 m to about 6 m. The motor vehicle includes a LIDAR sensor 4 , which detects an area on the side of the motor vehicle, for example a protective barrier. At the first location s1, the LIDAR sensor has a first detection area 6 and at the second location s2 a second detection area 8th on.

The car 2 detected by means of the LIDAR sensor 4 a first point cloud of which significant exemplary points 10 represented by squares in the figures. At the location s2 the motor vehicle is detected 2 by means of the LIDAR sensor a second point cloud, of the essential exemplary points by means of circles 12 are shown.

In 2 became the first point cloud 10 and the second point cloud 12 transformed into the world coordinate system. The first point cloud 10 was rotated clockwise about the center of the LIDAR sensor at location s1 in the world coordinate system. The second point cloud 12 became the center of the LIDAR sensor 4 turned clockwise at location s2. It can be seen that the point clouds do not match. Therefore, the cost function is not optimal.

In the presentation of 3 became the first point cloud 10 and the second point cloud 12 transformed into the world coordinate system. The first point cloud 10 became the center of the LIDAR sensor 4 turned counterclockwise at location s1, and the second point cloud 12 was rotated counterclockwise around the center of the LIDAR sensor in the world coordinate system at location s2. Once the first cloud of points 10 and the second point cloud 12 The first point cloud is rotated by the angle corresponding to the misalignment angle 10 and the second point cloud 12 partially in coverage, in particular in the area in which the at least one static object both at location s1 and at location s2 of the LIDAR sensor 4 was recorded ( 4 ). Consequently, the cost function for this angle is minimal.

The invention has the advantage that by means of an object located in the natural environment, for example a protective barrier, a maladjustment of a driver assistance sensor about the z-axis can be determined and such a misalignment can be compensated. Thus, the safety of an assistance system can be improved, also an even more reliable driver assistance sensor for fully autonomous driving of a motor vehicle is created.

One particular difficulty with a cloud of points generated by a LIDAR sensor is that, unlike radar data, there is no dynamic information of a detected target. With the inventive method for detecting a misalignment can be detected by means of reliable movement data without model knowledge and in a natural scene, a subsequent change in the installation position of the assistant sensor, wherein the Point clouds generated by the assistant sensors do not necessarily have to be further processed. However, detection and filtering of dynamic objects from the point clouds prior to application of the method has a positive effect on the quality of the result. The method according to the invention allows the driver assistance sensor to check itself. The misalignment can be estimated and compensated, so that a recalibration and a related workshop stay in many cases unnecessary.

In contrast to algorithms of the prior art, for example the ICP algorithm, the method according to the invention does not use a point cloud as a reference and tries to bring another cloud of points into the best possible agreement by means of rotation and translation with the reference point cloud. The method according to the invention turns both point clouds until a cost function becomes minimal. The method according to the invention severely restricts the search space of the algorithm, which leads to a significant reduction of the computational effort. This is possible because only a change in the orientation of the sensor is considered relevant during operation of the vehicle. A change in position is considered to be irrelevant because the position can only change due to a collision that is accompanied by a strong mechanical deformation of the vehicle, and anyway requires a workshop visit.

QUOTES INCLUDE IN THE DESCRIPTION

This list of the documents listed by the applicant has been generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Cited patent literature

• DE 102004056400 A [0007]

Claims (10)

Method for determining a misalignment of a driver assistance sensor ( 4 ) of a motor vehicle ( 2 ) with the following steps: - recording a first cloud of points ( 10 ) with the driver assistance sensor ( 4 ), where the first cloud of points ( 10 ) at least one stationary object in the vicinity of the motor vehicle ( 2 ) maps; - When the motor vehicle ( 2 ) has moved on a predetermined distance (s1-s2) or when a predetermined period of time has elapsed, picking up a second point cloud ( 12 ) with the driver assistance sensor ( 2 ), where the second cloud of points ( 12 ) the at least one stationary object in the environment of the motor vehicle ( 4 ) maps; - Transformation of the first cloud of points ( 10 ) and the second cloud of points ( 12 ) into a common coordinate system; - Stepwise rotation of the first point cloud ( 10 ) and the second cloud of points ( 12 ) in the same direction until an abort criterion is met; and - determining after each stepwise rotation of the first point cloud ( 10 ) and the second cloud of points ( 12 ) a measure of the agreement of the first point cloud ( 10 ) and the second cloud of points ( 12 ), wherein the abort criterion is met when the measure of correspondence has reached an optimal value; and - determining a rotation of the driver assistance sensor ( 4 ) about its z-axis based on the rotation angle of the first cloud of points ( 10 ) and the second point cloud ( 12 ), where the degree of agreement has reached an optimal value. The method of claim 1, characterized in that the step of determining the match comprises the step of minimizing a cost function. A method according to claim 2, characterized in that the step of minimizing the cost function comprises the step of minimizing the mean square distance. Method according to one of claims 1 to 3, characterized by the step of transforming the first cloud of points ( 10 ) and the second cloud of points ( 12 ) into the world coordinate system before the stepwise rotation of the first and the second cloud of points ( 10 . 12 ). Method according to one of claims 1 to 4, characterized in that the step of turning for the first cloud of points ( 10 ) a different center of rotation than for the second cloud of points ( 12 ) used. Method according to one of claims 1 to 5, characterized in that the driver assistance sensor (a LIDAR sensor 4 ). Method according to one of claims 1 to 6, characterized in that the at least one object is any stationary object in the vicinity of a roadway, the motor vehicle ( 2 ) during a ride with a driver from a starting point to a destination point, wherein the object is not for the adjustment of a driver assistance sensor ( 4 ), wherein the stationary object is preferably - a building, - a road boundary, - a protective barrier, - a guide post and / or - a plant. Method according to one of claims 1 to 7, characterized in that the steps of detecting the first cloud of points ( 10 ), the detection of the second cloud of points ( 12 ), the transformation into a common coordinate system and the determination of the rotation about the z-axis are repeatedly performed. Method according to claim 8, characterized by the step of filtering and / or averaging the detected rotations of the vehicle assistance sensor ( 4 ) around the z-axis.  Method according to one of claims 1 to 9, characterized by the step of compensating for the rotation of the driver assistance sensor about the z-axis.

Images