DE102019218059A1 - Detection of elongated objects - Google Patents

Abstract

The present invention relates to a device (16) for recognizing elongated objects (12) in the vicinity of a vehicle (14), having: an input interface (20) for receiving a radar target list of a radar sensor (18) with information on detected radar targets in the surroundings of the vehicle and movement data with information on a movement of the radar sensor in relation to the surroundings; an accumulation unit (22) for generating an accumulated radar target list in a sensor-fixed coordinate system based on the received radar target list and the movement data, with radar target lists of several time steps being combined with compensation for the movement of the radar sensor; a raster unit (24) for generating at least one raster representation (32) of the surroundings of the vehicle based on the accumulated radar target list, targets of the accumulated radar target list being assigned to raster cells of the raster representation which correspond to positions in the surroundings of the vehicle; and an evaluation unit (26) for determining a list of results with information on elongated objects in the vicinity of the vehicle based on an evaluation of the at least one raster representation. The present invention also relates to a method and a system (10) for recognizing elongated objects (12) in the vicinity of a vehicle (14).

Description

The present invention relates to a device for recognizing elongated objects in the surroundings of a vehicle. The present invention further relates to a system and a method for recognizing elongate objects in the surroundings of a vehicle.

Modern vehicles (cars, vans, trucks, motorcycles, etc.) have a large number of sensors that provide the driver with information and control individual functions of the vehicle in a partially or fully automated manner. The surroundings of the vehicle and other road users are recorded via sensors. Based on the recorded data, a model of the vehicle environment can be generated and changes in this vehicle environment can be reacted to.

Radar technology is an important sensor principle. Most of the radar sensors used today in the vehicle sector work as multi-pulse radar sensors (also referred to as chirp sequence radar sensors), in which several frequency-modulated pulses are transmitted at short intervals. The radar sensors typically include several transmitting and receiving elements (antenna array) that form virtual channels of the radar sensor (Rx / Tx antenna pairs). In each reception channel, the baseband signal is downmixed, filtered and then the baseband signal obtained in this way is digitized. By preprocessing the baseband signals for each receiving channel, a target, that is to say an object, can be detected and localized in a field of view of the radar sensor.

The preprocessing can in particular be a Fourier transformation or optimal filtering of the baseband data for a single radar pulse (so-called fast preprocessing, fast-time processing), a further Fourier transformation for several pulses (so-called slow preprocessing, slow-time processing), beamforming, signal power detection , for example based on constant force alarm rate method, as well as a high point determination. A radar target list (also referred to as a point cloud) can then be generated based on this preprocessed sensor data. This radar target list includes at least the target parameters distance, radial or Doppler speed and (if available) azimuth and elevation angles and forms the basis for an environment recognition.

A relevant problem in the field of autonomous and semi-autonomous driving concerns the detection of objects in the vicinity of a vehicle. Based on recognized objects, for example, one's own vehicle can be localized or a reaction of one's own vehicle can be adapted to the environment. The object detection based on radar sensor data is relevant on the one hand under ambient conditions in which cameras and other sensor technologies only work to a limited extent (rain, darkness, etc.) and on the other hand offers the advantage of a relatively large range. Relevant objects often have an elongated shape (lane boundaries, crash barriers, lanterns, traffic signs, lane lighting, tree trunks, ...) whereby particularly vertical, post-like, elongated objects can advantageously be used in self-localization.

In this context, Werber et al., "RoughCough - A New Image Registration Method for Radar Based Vehicle Self-Localization", 2015, describes a method for image registration that is suitable for use with radar raster maps. The method does not require any sharply structured input images and can be used for all image pairs that can be aligned by means of a Euclidean transformation. Few line-like objects in the reference image are used for the description. The similarity of a test image to these reference lines is determined in a fast Hough transformation.

The disadvantage of previous approaches is that a comparatively high computing power is often required for object recognition. Approaches in which different sensor principles are used and a large number of different object types are to be detected have a high demand for resources and are often error-prone.

Based on this, the present invention has the task of providing an approach for the radar sensor-based detection of elongated objects in the surroundings of a vehicle, which can be calculated efficiently and which enables a reliable detection of elongated objects. In particular, post-like, that is, vertically aligned and straight, elongated objects are to be recognized. Such post-like objects, but also other elongated objects, in particular line-like, that is to say straight elongated objects, allow a vehicle to be precisely located itself.

• einer Eingangsschnittstelle zum Empfangen einer Radar-Zieleliste eines Radarsensors mit Informationen zu detektierten Radarzielen in der Umgebung des Fahrzeugs und von Bewegungsdaten mit Informationen zu einer Bewegung des Radarsensors gegenüber der Umgebung;
• einer Akkumulationseinheit zum Erzeugen einer akkumulierten Radar-Zieleliste in einem sensorfesten Koordinatensystem basierend auf der empfangenen Radar-Zieleliste und den Bewegungsdaten, wobei Radar-Zielelisten mehrerer Zeitschritte unter Kompensation der Bewegung des Radarsensors zusammengefasst werden;
• einer Rastereinheit zum Erzeugen mindestens einer Rasterdarstellung der Umgebung des Fahrzeugs basierend auf der akkumulierten Radar-Zieleliste, wobei Ziele der akkumulierten Radar-Zieleliste Rasterzellen der Rasterdarstellung zugeordnet werden, die Positionen in der Umgebung des Fahrzeugs entsprechen; und
• einer Auswerteeinheit zum Ermitteln einer Ergebnisliste mit Angaben zu länglichen Objekten in der Umgebung des Fahrzeugs basierend auf einer Auswertung der mindestens einen Rasterdarstellung.

To achieve this object, the invention relates in a first aspect to a device for recognizing elongated objects in the surroundings of a vehicle, with:

• an input interface for receiving a radar target list of a radar sensor with information on detected radar targets in the surroundings of the vehicle and movement data with information on a movement of the radar sensor in relation to the surroundings;
• an accumulation unit for generating an accumulated radar target list in a sensor-fixed coordinate system based on the received radar target list and the movement data, with radar target lists of several time steps being combined with compensation for the movement of the radar sensor;
• a raster unit for generating at least one raster representation of the surroundings of the vehicle based on the accumulated radar target list, targets of the accumulated radar target list being assigned to raster cells of the raster representation which correspond to positions in the surroundings of the vehicle; and
• an evaluation unit for determining a list of results with information on elongated objects in the vicinity of the vehicle based on an evaluation of the at least one raster representation.

• einem Radarsensor zum Detektieren von Radarzielen in der Umgebung des Fahrzeugs, wobei der Radarsensor vorzugsweise als frequenzmodulierter Multipuls-Radarsensor ausgebildet ist; und
• einer Vorrichtung wie zuvor beschrieben.

In a further aspect, the present invention relates to a system for recognizing elongated objects in the surroundings of a vehicle, with:

• a radar sensor for detecting radar targets in the vicinity of the vehicle, wherein the radar sensor is preferably designed as a frequency-modulated multi-pulse radar sensor; and
• a device as described above.

Further aspects of the invention relate to a method embodied in accordance with the device described above and a computer program product with program code for performing the steps of the method when the program code is executed on a computer, as well as a storage medium on which a computer program is stored which, when it is on a computer is executed, causes an execution of the method described herein.

Preferred embodiments of the invention are described in the dependent claims. It goes without saying that the features mentioned above and those yet to be explained below can be used not only in the respectively specified combination, but also in other combinations or on their own, without departing from the scope of the present invention. In particular, the system, the method and the computer program product can be implemented in accordance with the configurations described for the device in the dependent claims.

According to the invention, it is provided that data from a radar sensor are received via an input interface. In particular, a radar target list is received which contains various targets detected by means of a radar sensor. The radar target list contains various target parameters for each target and is usually made available by the radar sensor at regular intervals. In the device according to the invention, an accumulated radar target list is generated which summarizes the received radar target lists over several time steps. This can increase the resolution. In the summary, a movement of the radar sensor between the time of generating a first radar target list and the time of generating a second radar target list is taken into account or compensated based on the movement data. Corresponding transformations are used for this purpose in order to transform the radar target lists of different detection times into a uniform, sensor-fixed coordinate system. Based on the accumulated radar target list, a raster representation of the surroundings of the vehicle is generated. In particular, a two-dimensional raster representation is generated. The radar targets are assigned to individual cells in a grid. The surroundings of the vehicle are thus discretized. On the basis of this raster representation, a list of results is generated that includes information on elongated objects in the vicinity of the vehicle. Exactly one grid display can be used here. It is also possible to use several raster representations based on different parameters.

Compared to previous approaches, in which a direct evaluation of the radar target list is provided without accumulation over several time steps, the approach according to the invention enables an improved resolution and thus a more precise and reliable detection. By using a raster representation, efficient calculability can be achieved. Compared to a direct application of an object recognition algorithm to the radar target list, the use of the raster display results in a significantly improved calculability. This makes it possible to realize a reliable and precise detection of elongated objects.

In a preferred embodiment, the input interface is designed to receive the radar target list with an indication of a radar cross section for a radar target. The raster unit is designed to generate a two-dimensional radar cross-section raster representation. A raster cell of the radar cross-section raster display includes an indication of a maximum radar cross-section, an average radar cross-section, a quantile of a radar cross-section and / or a profile of a radar cross-section in a vertical direction. A radar cross-section raster representation can also be referred to as a Radar Cross Section (RCS) grid. The radar cross-section of a reflection or a target indicates how strong the target's reflections are in the direction of the radar sensor. In particular in the case of metallic post-like objects with a round cross-section (traffic signs), reliable object detection can be carried out based on the radar cross-section. Various parameters of the radar cross-section can be used. The result is a reliable detection of perpendicular objects with a round cross-section.

In a preferred embodiment, the raster unit is designed to generate a two-dimensional multipath raster representation. A grid cell of the multipath grid representation comprises a minimum of a vertical object extent above and below an estimated subsurface plane. It is analyzed whether an object has a comparable size both in the direct reflection and in an indirect reflection over a subsurface plane. Highly reflective objects, such as, for example, vertical metallic post-like objects with a round cross-section, have an almost equally strong reflection on the ground plane. In this respect, through the use of the multipath raster representation, a reliable object recognition of perpendicularly elongated objects can also take place.

In a preferred embodiment, the raster unit is designed to generate a two-dimensional weighted evidence raster display. A grid cell of the weighted evidence grid representation comprises a ratio between a number of radar targets at different heights at a two-dimensional position corresponding to the grid cell and an expected number of radar targets of a vertical elongated object. The raster unit is designed to precompute the expected number of radar targets of the vertical elongated object based on a height expansion angle determined from the cumulative radar target list and an angular resolution of the radar sensor. A weighted evidence raster representation corresponds to a representation in which a relationship between a number of expected detections for a vertical elongated object and a number of actual detections is evaluated. Vertically standing elongated objects (post-like objects) show a high number of detections. By taking the ratio into account, a high level of reliability can be achieved in the object recognition of a corresponding object. In particular, it is possible to identify vertical metallic objects with a round cross-section (street lamps, traffic signs, etc.).

In a preferred embodiment, the raster unit is designed to generate a two-dimensional evidence raster display. A grid cell of the evidence grid display comprises a number of radar targets at different heights at a two-dimensional position corresponding to the grid cell. An evidence grid display (also referred to as an evidence grid) indicates the number of detections for a specific two-dimensional position. Two-dimensional positions with a higher number of detections or radar targets are occupied by objects. The use of a two-dimensional evidence raster representation makes it possible to apply image processing algorithms to this two-dimensional representation, which enable an efficient detection of line-like objects that run horizontally.

In a preferred embodiment, the evaluation unit is designed to determine a list of results with information on vertical elongated objects based on a high point detection in the at least one raster representation. Alternatively or additionally it is possible that a decision tree algorithm is used. Furthermore, as an alternative or in addition, it is possible that a pre-trained learning algorithm is used. In particular, vertical elongated objects can be identified. It is particularly advantageous if post-like, vertical elongated objects are identified which have a round cross section and are made of metal. Such objects can be, for example, parts of street lamps, traffic signs or lane boundaries. Such objects are mostly stationary and can advantageously be used for self-localization. Such objects can be recognized with high reliability.

In a preferred embodiment, the evaluation unit is designed to determine a list of results with information on horizontal elongated objects based on an application of a computer vision approach to the at least one raster display. Using a computer vision approach can ensure efficient computation. A high update rate can be implemented. Elongated objects running horizontally can in particular run parallel to a ground plane. For example, crash barriers or other road boundaries can be recognized with high reliability.

In a preferred embodiment, the evaluation unit is designed to assign the elongated objects to predefined object classes. In addition to the object class, the result list preferably includes an assignment confidence which indicates the reliability of the assignment. Due to the classification, a dem response of the autonomous or semi-autonomous vehicle adapted to the respective object. It is also possible that the recognized object classes are used for further processing in the environment recognition. If an assignment confidence is also calculated, an adjustment can be made during further processing. The reliability can be improved.

In a preferred embodiment, the evaluation unit is designed to assign the elongated objects to predefined object classes based on a two-dimensional position of the recognized elongated object in relation to a probable trajectory of the radar sensor. Additionally or alternatively, the evaluation unit is designed to assign the elongated objects to predefined object classes based on a classifier, which is preferably pre-trained based on lidar data from a lidar sensor. It is possible for the position of a recognized elongate object to be used in the object classification in comparison to a probable future trajectory of the radar sensor or of the vehicle. It can be assumed here, for example, that objects which are in the immediate vicinity of the roadway or the probable trajectory are more likely to be roadway boundaries. In addition, assumptions can be made with regard to the arrangement of traffic signs or similar objects and taken into account in the classification. The classifier can preferably be pre-trained based on data from another sensor. In particular, lidar data can be used that enable highly precise object recognition.

In a preferred embodiment, the input interface is designed to receive camera data from a camera with information on elongated objects in the vicinity of the vehicle. The evaluation unit is designed to assign the elongated objects to predefined object classes based on semantic segmentation using the camera data. It is possible for additional data from a camera to be taken into account in order to improve the classification.

In particular, semantic segmentation can be used for this, which enables reliable object recognition.

In a preferred embodiment, the evaluation unit is designed to determine a recognition confidence for an elongated object, which indicates the reliability of the recognition. In addition, the evaluation unit is designed to remove recognized elongated objects from the list of results which are located within a predefined immediate vicinity of an elongated object with a higher recognition confidence. The result list has been revised, in which objects are sorted according to the reliability of their recognition. Objects that are in the immediate vicinity of another object are removed from the result list. As a result, duplicate objects or incorrect detections or duplicate detections can be recognized and avoided. An object is shown as precisely as possible in the results list.

In a preferred embodiment, the input interface is designed to receive the movement data from a vehicle odometry system and / or from a position sensor. In particular, an inertial sensor-based system can be used as the vehicle odometry system. It is also possible that dead reckoning approaches are used. The vehicle odometry system outputs information about a movement of the vehicle between two time steps. This information can then be used to transform the radar target lists generated at the two points in time into a coordinate system that is fixed to the sensor or vehicle. It is also possible for data from a position sensor to be used. The result is a reliable detection of the movement of the vehicle and an accurate transformation in the generation of the accumulated radar target list.

In a preferred embodiment, the raster unit is designed to apply a two-dimensional filter, in particular a Gaussian filter, to the generated at least one raster representation. The use of a Gaussian filter can be used for smoothing. In subsequent processing steps, object recognition can be carried out with higher accuracy.

An elongated object can in particular be a straight line or post-like object. For example, (vertical) posts of traffic signs or of lane boundaries or also (horizontally running) crash barriers can represent elongated objects. A horizontally elongated object is understood to mean an object that runs essentially horizontally to a subsurface. A vertical elongated object is understood to mean an object that is essentially perpendicular to the underground level (street level). In particular, objects can be recognized.

A radar sensor sends out a radar signal and receives reflections of the radar signal from objects (also known as targets) within a field of view of the radar sensor. The field of view describes an area within which objects can be detected. A radar sensor can comprise several individual sensors, which, for example, have a 360 ° Enabling all-round visibility and thus being able to record a complete image of the surroundings of a vehicle. A radar target list can also be referred to as a point cloud and corresponds to a list of detections of this radar sensor output by a current radar sensor. The list includes several parameters for a single detection. In particular, the radar target list can include angle information (azimuth, elevation) as well as a distance (range), a radial or Doppler speed and also a radar cross section.

A grid display is understood to mean a display as a grid, in which different radar targets or scan points of a radar sensor are assigned to (discrete) grid cells of the grid on the basis of their parameters. Basically, a raster representation can be two or three dimensional. A two-dimensional grid running parallel to a floor plane or to a horizontal plane of the vehicle is preferably used here as the raster representation. A raster representation is understood to mean modeling of the surroundings of a vehicle. Each grid cell contains information about the detection of objects within the area of the surroundings of the vehicle represented by the grid cell. In this respect, a raster representation corresponds to a discretization of the radar target list. In particular, an area in front of the vehicle can be mapped through the grid display. A list of results contains the recognized elongated objects. In particular, parameters such as position or type and confidence can be contained in the result list for the objects.

• 1 eine schematische Darstellung eines erfindungsgemäßen Systems;
• 2 eine schematische Darstellung einer erfindungsgemäßen Vorrichtung;
• 3 eine schematische Darstellung einer Darstellung verschiedener Radarziele in einer Evidenz-Rasterdarstellung;
• 4 eine schematische Darstellung eines Reflexionspfads über die Bodenebene;
• 5 eine schematische Darstellung eines weiteren Reflexionspfads über die Bodenebene für ein höheres Objekt;
• 6 eine schematische Darstellung einer Detektion von pfostenartigen Objekten nach der vorliegenden Erfindung;
• 7 eine schematische Darstellung eines Fahrzeugs mit einem erfindungsgemäßen System; und
• 8 eine schematische Darstellung eines erfindungsgemäßen Verfahrens.

The invention is described and explained in more detail below using a few selected exemplary embodiments in connection with the accompanying drawings. Show it:

• 1 a schematic representation of a system according to the invention;
• 2 a schematic representation of a device according to the invention;
• 3rd a schematic representation of a representation of various radar targets in an evidence raster representation;
• 4th a schematic representation of a reflection path over the ground plane;
• 5 a schematic representation of a further reflection path over the ground plane for a higher object;
• 6th a schematic representation of a detection of post-like objects according to the present invention;
• 7th a schematic representation of a vehicle with a system according to the invention; and
• 8th a schematic representation of a method according to the invention.

In the 1 is schematically a system according to the invention 10 for recognizing elongated objects 12th in an environment of a vehicle 14th shown. The system 10 includes a device 16 as well as a radar sensor 18th , which in the illustrated embodiment both in the vehicle 14th are integrated. Via the radar sensor 18th become different objects around the vehicle 14th detected. In the device 16 elongated objects are created based on the radar target lists that the radar sensor generates at different times 12th in the vicinity of the vehicle 14th recognized. In particular, in the device 16 generates a list of results that can be used in downstream processing steps to adapt a reaction of an autonomous or semi-autonomous vehicle to ambient conditions or to localize the vehicle 14th in relation to the environment.

The approach of the present invention relates in particular to the detection of elongated objects. On the one hand, elongated objects running vertically, such as posts or post-like objects, can be recognized. On the other hand, horizontally extending elongated objects such as guardrails or the like can be recognized. In particular, it is advantageously possible according to the invention to recognize metallic objects with a round cross section that are perpendicular. For example, the feet of traffic signs or traffic lights can be recognized with high reliability.

In the 2 is schematically a device according to the invention 16 shown. The device 16 includes an input interface 20th , a unit of accumulation 22nd , a grid unit 24 and an evaluation unit 26th . The device according to the invention can, for example, be integrated into a vehicle control device or be designed as part of a driver assistance system or also be implemented as a separate module. It is possible that the device according to the invention 16 is partially or completely implemented in software and / or hardware. The various units and interfaces can be designed individually or in combination as a processor, processor module or software for a processor.

Via the input interface 20th radar target lists from a radar sensor are received. The input interface 20th be connected to a vehicle bus system, for example. The input interface 20th can as Plug connection can be implemented in hardware or designed as a software interface for receiving data. The received radar target lists contain information on detected radar targets in the vicinity of the vehicle. In particular, the radar target lists can be received by a radar sensor which is attached to a vehicle in the area of a front bumper. Furthermore, movement data with information on a movement of the radar sensor in relation to the environment are received via the input interface. In particular, movement data can be received with information about movement of the vehicle. For example, the input interface 20th be connected to a vehicle odometry system. It is also possible that the input interface 20th is connected to a corresponding position sensor (GPS, GLONASS, Galileo etc.).

In the accumulation unit 22nd the radar target lists are summarized over time. To this extent, several radar target lists are combined in a sensor-fixed or vehicle-fixed coordinate system. This allows an increased resolution to be achieved. A sensor-fixed coordinate system is fixed with regard to the orientation and position in relation to the radar sensor or the vehicle. The sensor-fixed coordinate system is an image of the environment as it is observed by the sensor. On the one hand, the vehicle and thus a radar sensor attached to the own vehicle moves between the detections of the radar sensor, so that the relative position of static targets changes. On the other hand, the surface of the object hit by the radar signal can also change, so that other angles of incidence of the reflected radar signals result.

In order to be able to accumulate radar target lists of the radar sensor that were generated at several successive points in time, it is necessary to carry out a corresponding (coordinate) transformation. After each step of updating the accumulated radar target list, a transformation into the raster display is carried out based on the selected type of raster. The corresponding coordinate transformation can in particular be expressed by a translation matrix and a rotation matrix. The coordinate transformation can, for example, be calculated on the basis of the speed of one's own vehicle and a measured rate of rotation. It is also possible to use a GPS system or another motion estimation system. A three-dimensional speed vector of the vehicle can be used to compensate for the movement or for the transformation 14th as well as three-dimensional angular rotation rates of the vehicle 14th used (movement data). These data can be determined using appropriate motion estimation methods. The accumulated radar target list generated is based, for example, on a predefined number of the last radar targets. Older radar targets are removed from the accumulated radar target list. The accumulation virtually improves the resolution of the radar sensor used.

In the grid unit 24 at least one raster representation of the surroundings of the vehicle is generated based on the accumulated radar target list. For this purpose, the radar target list (also referred to as a point cloud) is discretized. A discretization is understood to mean an assignment of the individual entries of the accumulated radar target list to cells of a predefined grid. For example, a grid with a cell size of half a meter can be used. In particular, a two-dimensional grid can be generated.

A two-dimensional filter (for example a Gaussian filter) can be used in the preprocessing of the raster representation in order to achieve a denser image representation. This can improve the detection performance of a corresponding computer vision approach.

In the evaluation unit 26th a result list is generated based on this at least one raster representation, which includes information on elongated objects in the vicinity of the vehicle. The evaluation unit 26th be designed in particular to apply an algorithm of image processing or computer vision (computer image processing) to the raster display. In particular, it is possible for a post and / or line extraction to be carried out.

For the detection of posts (vertically aligned elongated objects), for example, an approach such as the Adaptive and Generic Accelerated Segment Test (AGAST) of the Technical University of Munich can be used, which can be applied to a signal-to-noise ratio raster display.

Various line recognition algorithms of computer vision can be used to extract elongated objects that run vertically and essentially parallel to the ground. For example, the following algorithms can be used for this purpose: Hough, Iterative Hough, LSD. It is also possible that machine learning approaches are used. The evaluation unit 26th can then transmit the ascertained list of results to a corresponding vehicle control device, which then makes the information available to the vehicle driver or carries out corresponding further processing.

It is possible to classify the detected elongated objects. For example, a zone-based concept can be used for this purpose. Lines or other elongated objects that are in the vicinity of a probable trajectory of the own vehicle are very likely to be the lane boundaries. Other lines or elongated objects are most likely parts of buildings or other structural elements.

In the result list, features or lines with a corresponding reference value in the two-dimensional coordinate system fixed to the sensor or vehicle can be recorded together with the associated properties such as the radius for a vertical elongated object (post). For a vertical elongated object, for example, a polynomial representation or a point can be saved together with a length. In addition, uncertainties can be added both for the reference point and for the corresponding properties.

It is possible to access other sensor technologies for the classification. For example, a camera-based semantic segmentation based on the detected features can be carried out downstream in order to achieve a precise classification. For this purpose, the detected elongated objects can be projected into a camera frame in order to apply semantic segmentation.

It is also possible to access machine learning approaches. Based on this, elongated objects and in particular lines and posts can be detected with higher accuracy and with higher calculation efficiency. For example, detected lines or posts of a high-resolution lidar system can be used as training data.

In the 3rd the generation of a two-dimensional evidence raster display is shown schematically as an example. The three representations are a bird's-eye view of the vehicle 14th and an area in front of the vehicle. In the field of vision 28 a radar sensor on the vehicle 14th there is an object 13th , which in turn has multiple radar targets 30th are located. Radar target lists from the radar sensor are received at a number of successive points in time, each of which includes information about the detected radar targets. In the middle representation is a raster representation 32 indicated in Cartesian coordinates, which includes an area in front of the vehicle. The object 12th lies within the grid display. As indicated on the right-hand side, in the example shown, the number of radar targets within the cell is entered in the individual cells of the grid display.

Alternatively, it would also be possible in a grid display 32 Enter minimum values, maximum values, median values, mean values of a signal-to-noise ratio (SNR grid) or a radar cross section (RCS grid).

In the 4th is an example of the effect of a reflection on the ground on a received radar signal. A relevant difference between radar and other active distance measurement methods is the comparatively high sensitivity with regard to multipath propagation. Multipath propagation means that the received signal does not come directly from a reflection on an object, but was reflected between the object and the radar sensor on another object or on a subsurface (ground level). Multipath propagation or reflections on the ground and especially on the road surface are particularly relevant. In this case, the reflections of the object are perceived in the same way as the multipath propagation. This can increase the measured distance slightly.

In the 4th is schematically a radar sensor 18th shown. With the radar sensor 18th becomes an object 13th detected at a distance of R. The radar sensor 18th is at a height of h above ground level 36 appropriate. In addition to direct reflection on the object 13th there will also be multipath propagation via a reflection in the ground plane 36 detected. The distance d results from this $$d=\sqrt{\frac{{\mathrm{R.}}^2}{\mathrm{4th}}+{\mathrm{<figure-callout\; id="2"\; label="H"\; filenames="DE102019218059A1\_0007.png"\; state="\{\{state\}\}">H</figure-callout>}}^2}$$

This gives the indirect distance r _i zu $$r_i=\frac{\mathrm{R.}+2\mathrm{<figure-callout\; id="2"\; label="d"\; filenames="DE102019218059A1\_0007.png"\; state="\{\{state\}\}">d</figure-callout>}}{2}=\frac{\mathrm{R.}}{2}+\sqrt{\frac{{\mathrm{R.}}^2}{\mathrm{4th}}+{\mathrm{<figure-callout\; id="2"\; label="H"\; filenames="DE102019218059A1\_0007.png"\; state="\{\{state\}\}">H</figure-callout>}}^2}$$

Depending on the height h of the radar sensor 18th above ground level 36 is appropriate, there are relevant differences between the true distance R and the measured indirect distance r _i .

der indirekte Abstand r_i $$r_i=\frac{\left(1+\frac{p}{h}\right)\sqrt{{\left(\frac{R}{1+\frac{p}{h}}\right)}^2+h^2+\sqrt{R^2+{\left(p-h\right)}^2}}}{2}$$

basierend auf dem direkten Abstand r_d $$r_d=\sqrt{R^2+{\left(p-h\right)}^2}$$

In the 5 is a schematic of the detection of a post-like object with a height p by means of the radar sensor 18th shown. Through the reflection on the ground plane 36 results for the indirect distance r _i with the distances d1 and d2 starting from $${\mathrm{<figure-callout\; id="1"\; label="d"\; filenames="DE102019218059A1\_0007.png"\; state="\{\{state\}\}">d</figure-callout>}}_1+{\mathrm{<figure-callout\; id="2"\; label="d"\; filenames="DE102019218059A1\_0007.png"\; state="\{\{state\}\}">d</figure-callout>}}_2=\left(1+\frac{p}{H}\right)\sqrt{{\left(\frac{\mathrm{R.}}{1+\frac{p}{H}}\right)}^2+{\mathrm{<figure-callout\; id="2"\; label="H"\; filenames="DE102019218059A1\_0007.png"\; state="\{\{state\}\}">H</figure-callout>}}^2}$$

the indirect distance r _i $$r_i=\frac{\left(1+\frac{p}{H}\right)\sqrt{{\left(\frac{\mathrm{R.}}{1+\frac{p}{H}}\right)}^2+{\mathrm{<figure-callout\; id="2"\; label="H"\; filenames="DE102019218059A1\_0007.png"\; state="\{\{state\}\}">H</figure-callout>}}^2+\sqrt{{\mathrm{R.}}^2+{\left(p-H\right)}^2}}}{2}$$

based on the direct distance r _d $$r_d=\sqrt{{\mathrm{R.}}^2+{\left(p-H\right)}^2}$$

Based on these calculations, it results that a maximum distance at an upper tip T of the object 13th , if it is a vertical elongated object, for large distances between radar sensors 18th and object 13th is comparatively low. As a result, it can approximately be assumed that in the vast majority of cases both the tip T of a vertical elongated object and the rest of the vertical elongated object fall into the same grid cell, taking into account the multipath propagation.

In order to identify a vertical elongated object, it is possible that a simple model is used to set up an expected number of radar targets that an elongated object with a certain height causes at a previously known angular resolution of the sensor. The expected number depends on the distance between the radar sensor and the elongated object. First, an elevation angle is calculated. This elevation angle can then be divided by the angular resolution of the radar sensor in order to obtain a number of expected radar targets for a measurement at this given distance.

Additionally or alternatively, it is possible to evaluate a radar cross-section in a corresponding grid display. The radar cross-sections of targets for individual reflections on a post-like object, in particular on a metallic object such as a bracket for a street lamp or a traffic sign, is characteristic. This characteristic can be used to differentiate between different types or classes of post-like objects. For an ideal post-like object, there is a maximum expected reflection at the height at which the radar sensor is installed. Other reflections at higher or lower portions of the post-like object have a smaller radar cross-section. Traffic signs or street lamps or other objects attached to the top of a post-like object require specific reflections at the end of the post-like object. It is possible to generate a radar cross-sectional profile across the height of the post-like object by associating between radar measurements within a cylindrical area around the object.

In the 6th an inventive approach for recognizing elongated vertical objects (post-like objects) is shown schematically. First, several raster displays are generated. In the example shown in the 6th a weighted evidence grid (Expectancy Weighted Evidence Grid, WEG), a multipath grid (MG) and a radar cross section grid (RCSG) are generated. The alignment of the two-dimensional raster representations WEG, MG, RCSG is parallel to the ground plane. It is assumed that the elongated vertical objects (posts) to be detected are oriented perpendicular to the ground plane. The x-axis of the raster representations WEG, MG, RCSG is oriented in the direction of movement of the vehicle or of the radar sensor. The y-axis runs orthogonally to this and parallel to the ground plane. Each measurement by the radar sensor is assigned to a grid cell of a corresponding x and y position.

The evaluation of the weighted evidence grid representation WEG is based on the fact that a post-like, vertical elongated object forms a good reflector from all sides due to its geometry. Based on the angular resolution of the radar sensor in the elevation direction, a number of expected detections for a specific post-like object with a previously known height at a specific distance can be derived. In a normal evidence grid display, the value in the corresponding grid cell would only be increased by 1 for a measurement within this grid cell. In the weighted evidence grid display proposed according to the invention, weighting is carried out based on the expected number of detections.

The multipath raster representation MG is based on the characteristic multipath propagation over the ground plane of a post-like vertical elongated object. The multipath propagation over the ground level only leads to a slight increase in the measured distance. Therefore, a minimum between a height above and below the ground level can be used as an indicator of the Multipath can be used to identify a post-like object.

The radar cross-section raster display RCSG can be used to set up various indicators based on the radar cross-section values of the radar targets assigned to a raster cell. In particular, maximum values, mean values and quantiles of the radar cross-sections of the measurements that fall into the grid cell can be entered. In addition, it is possible that a radar cross-sectional profile is calculated along the height axis of the post-like vertical elongated object. For this purpose, the post-like object is divided into smaller sections and a mean radar cross-section is calculated for each section. This results in a radar cross-sectional profile in the form of a function of the radar cross-section over the height of a post-like elongate object.

A high point detection PD can be applied to the multipath raster representation MG and the weighted evidence raster representation WEG. A corresponding filter method can be used for this purpose. For example a Laplace transform or a Gaussian filter. In order to use such a filter, an order of magnitude of the high point to be detected must be known. For known post-like, vertical elongated objects such as traffic signs or street lights, a previously known scaling can be used for this purpose. An adjustment can be made in order to detect other objects such as trees or advertising pillars.

There are various candidates that are summarized in a results list PC.

In further processing steps S1, S2, S3, the result list PC can then be sorted on the basis of a further indicator value. In particular, in a sorting step S1, sorting according to descending radar cross-section values or entries in the radar cross-section raster display can be carried out. In addition to this sorting, a local neighborhood of the elongated objects identified is examined in a second processing step S2. The detected elongated objects in a local neighborhood with a predefined radius of objects with a higher radar cross-section are deleted in order to rule out multiple detections of the same object. In the last processing step S3 there is a reduced list of results with information on post-like elongated objects in the vicinity of the vehicle.

In the 7th is schematically a vehicle according to the invention 14th shown. The vehicle 14th has a radar sensor 18th and a device according to the invention 16 on. In the example shown, the vehicle 14th furthermore a position sensor 38 on, which can be attached, for example, to the roof of the vehicle. In addition, the vehicle shown has 14th an inertial measuring unit 40 which can be used to provide movement data with information on the movement of the vehicle 14th or the radar sensor attached to the vehicle 18th to determine.

In the 8th a method according to the invention for recognizing elongated objects in the surroundings of a vehicle is shown schematically. The method comprises the steps of receiving S10 a radar target list and movement data, generating S12 an accumulated radar target list, generating S14 at least one raster representation and determining S16 a list of results. The method can in particular be implemented in software that is executed on a vehicle control device. It is also possible for the method to be carried out in a radar sensor.

The invention has been comprehensively described and explained with reference to the drawings and the description. The description and explanation are to be understood as examples and not restrictive. The invention is not limited to the disclosed embodiments. Other embodiments or variations will become apparent to those skilled in the art after using the present invention and after carefully analyzing the drawings, the disclosure and the following claims.

In the claims, the words “comprising” and “having” do not exclude the presence of further elements or steps. The undefined article “a” or “an” does not exclude the presence of a plural. A single element or a single unit can perform the functions of several of the units mentioned in the patent claims. An element, a unit, an interface, a device and a system can be implemented partially or completely in hardware and / or in software. The mere mention of some measures in several different dependent patent claims should not be understood to mean that a combination of these measures cannot also be used advantageously. A computer program can be stored / distributed on a non-volatile data carrier, for example on an optical memory or on a semiconductor drive (SSD). A computer program can be distributed together with hardware and / or as part of hardware, for example by means of the Internet or by means of wired or wireless communication systems. Reference signs in the patent claims are not to be understood as restrictive.

List of reference symbols

10

system

12th

elongated object

13th

object

14th

vehicle

16

contraption

18th

Radar sensor

20th

Input interface

22nd

Accumulation unit

24

Grid unit

26th

Evaluation unit

28

Field of view

30th

Radar target

32

Raster display

36

Ground level

38

Position sensor

40

Inertial measurement unit

Claims (15)

Device (16) for recognizing elongated objects (12) in the surroundings of a vehicle (14), with: an input interface (20) for receiving a radar target list of a radar sensor (18) with information on detected radar targets (30) in the vicinity of the vehicle and movement data with information on a movement of the radar sensor in relation to the environment; an accumulation unit (22) for generating an accumulated radar target list in a sensor-fixed coordinate system based on the received radar target list and the movement data, with radar target lists of several time steps being combined with compensation for the movement of the radar sensor; a raster unit (24) for generating at least one raster representation (32) of the surroundings of the vehicle based on the accumulated radar target list, targets of the accumulated radar target list being assigned to raster cells of the raster representation which correspond to positions in the surroundings of the vehicle; and an evaluation unit (26) for determining a list of results with information on elongated objects in the vicinity of the vehicle based on an evaluation of the at least one raster representation. Device (16) after Claim 1 wherein the input interface (20) is designed to receive the radar target list with an indication of a radar cross section for a radar target (30); the raster unit (24) is designed to generate a two-dimensional radar cross-section raster representation (32); and a raster cell of the radar cross-section raster display includes an indication of a maximum radar cross-section, an average radar cross-section, a quantile of a radar cross-section and / or a profile of a radar cross-section in a vertical direction. Device (16) according to one of the preceding claims, wherein the raster unit (24) is configured to generate a two-dimensional multipath raster representation (32); and a grid cell of the multipath grid representation comprises a minimum of a vertical object extent above and below an estimated subsurface plane. Device (16) according to one of the preceding claims, wherein the raster unit (24) is designed to generate a two-dimensional weighted evidence raster representation; a grid cell of the weighted evidence grid representation comprises a ratio between a number of radar targets (30) at different heights at a two-dimensional position corresponding to the grid cell and an expected number of radar targets of a vertical elongated object; and the raster unit is designed to precalculate the expected number of radar targets of the vertical elongated object based on a height expansion angle determined from the accumulated radar target list and an angular resolution of the radar sensor (18). Device (16) according to one of the preceding claims, wherein the raster unit (24) is designed to generate a two-dimensional evidence raster representation; and a grid cell of the evidence grid display comprises a number of radar targets (30) at different heights at a two-dimensional position corresponding to the grid cell. Device (16) according to one of the preceding claims, wherein the evaluation unit (26) for Determining a result list with information on vertical elongated objects (12) based on a high point detection in which at least one raster representation (32) is formed; a decision tree algorithm is formed; and / or a pre-trained learning algorithm. Device (16) according to one of the preceding claims, wherein the evaluation unit (26) is designed to determine a result list with information on horizontal elongated objects (12) based on an application of a computer vision approach to the at least one raster representation (32). Device (16) according to one of the preceding claims, wherein the evaluation unit (26) is designed to assign the elongated objects (12) to predefined object classes; and the result list, in addition to the object class, preferably includes an assignment confidence which indicates a reliability of the assignment. Device (16) after Claim 8 wherein the evaluation unit (26) is designed to assign the elongated objects (12) to predefined object classes based on a two-dimensional position of the recognized elongated object in relation to a probable trajectory of the radar sensor (18); and / or a classifier is formed, which is preferably pre-trained based on lidar data of a lidar sensor. Device (16) according to one of the Claims 8 and 9 wherein the input interface (20) is designed to receive camera data from a camera with information on elongated objects (12) in the vicinity of the vehicle (14); and the evaluation unit (26) is designed to assign the elongated objects to predefined object classes based on semantic segmentation using the camera data. Device (16) according to one of the preceding claims, wherein the evaluation unit (26) is designed to determine a recognition confidence for an elongated object (12) which indicates a reliability of the recognition; and is designed to remove recognized elongated objects from the list of results which are located within a predefined immediate neighborhood of an elongated object with a higher recognition confidence. Device (16) according to one of the preceding claims, wherein the input interface (20) is designed to receive the movement data from a vehicle odometry system and / or from a position sensor. System (10) for recognizing elongated objects (12) in the surroundings of a vehicle (14), comprising: a radar sensor (18) for detecting radar targets (30) in the vicinity of the vehicle, the radar sensor preferably being designed as a frequency-modulated multi-pulse radar sensor; and a device (16) according to any one of the preceding claims. Method for recognizing elongated objects (12) in the vicinity of a vehicle (14), comprising the steps: Receiving (S10) a radar target list of a radar sensor (18) with information on detected radar targets (30) in the vicinity of the vehicle and movement data with information on a movement of the radar sensor in relation to the environment; Generating (S12) an accumulated radar target list in a sensor-fixed coordinate system based on the received radar target list and the movement data, with radar target lists of several time steps being combined with compensation for the movement of the radar sensor; Generating (S14) at least one grid representation (32) of the surroundings of the vehicle based on the accumulated radar target list, targets of the accumulated radar target list being assigned grid cells of the grid display which correspond to positions in the surroundings of the vehicle; and Determination (S16) of a list of results with information on elongated objects in the vicinity of the vehicle based on an evaluation of the at least one raster representation. Computer program product with program code for performing the steps of the method according to Claim 14 when the program code is executed on a computer.

Images