Method for detecting objects from point cloud data based on the orientations of vectors between neighboring points

Abstract

A method for detecting objects (14) from data of a point cloud (38), in particular a LiDAR point cloud (38), is described. Each point (40, 40F, 40N, 40R) of the point cloud (38) has at least one spatial coordinate (x, y, z) which specifies the position of the point (40, 40F, 40N, 40R) in a coordinate system (42) assigned to the point cloud (38). In the method, at least one orientation of a vector (64) between a viewing location (46) within the coordinate system (42) and the at least one point (40F) of the point cloud (38) is determined. At least one point (40F) of the point cloud (38) is selected as the focus point (40F).For the at least one focus point (40F), at least one orientation value is determined, which specifies at least one relative orientation of a focus vector (64) extending between the viewing location (46) and the at least one focus point (40F) to at least one focus neighbor vector (68) extending between the at least one focus point (40F) and at least one point (40) of the point cloud (38) selected as a neighbor point (40N) to the at least one focus point (40F). Based on at least two orientation values ​​(60, 72), contour data are determined which describe a contour of at least one object (14).

Description

Technical field The invention relates to a method for detecting objects from data of a point cloud, in particular a LiDAR point cloud, wherein each point of the point cloud has at least one spatial coordinate which indicates the position of the point in a coordinate system assigned to the point cloud, wherein in the method at least one orientation of a vector between a viewing location within the coordinate system and the at least one point is determined for at least one point of the point cloud. Furthermore, the invention relates to a data processing device, in particular a data processing device for a LiDAR system, especially for a vehicle, with at least part of means for carrying out a method for recognizing objects from point cloud data, in particular LiDAR point clouds. Furthermore, the invention relates to a detection system, in particular a LiDAR system, in particular a detection system for a vehicle, with at least one transmitting device for sending scanning signals into a monitoring area, at least one receiving device for receiving scanning signals reflected in the monitoring area as echo signals, and at least one data processing device, in particular a data processing device according to the invention, with at least one part of means for carrying out a method for detecting objects from data of point clouds, in particular of LiDAR point clouds. Furthermore, the invention relates to a driver assistance system with at least part of means for carrying out a method for recognizing objects from point cloud data, in particular LiDAR point clouds. Furthermore, the invention relates to a vehicle with at least one detection system, in particular at least one LiDAR system, for monitoring a monitoring area in the vicinity of the vehicle and with at least part of means for carrying out a method for detecting objects from point cloud data, in particular LiDAR point clouds, which are generated with the at least one detection system. State of the art From US patent 2021 / 0272305 A1, a method for determining spatial relationships of point cloud clusters in a device comprising one or more processors and non-transitory memory is known. The method comprises obtaining a point cloud containing a plurality of points, wherein the plurality of points includes a first cluster of points and a second cluster of points, a specific point of the first cluster of points being associated with a characterization vector comprising a set of coordinates of the specific point in three-dimensional space and a cluster identifier of the first cluster of points. The method includes determining a spatial relationship vector based on the volumetric arrangement of the first and second point clusters, the spatial relationship vector characterizing the spatial relationship between the first and second point clusters.The procedure involves adding the spatial relationship vector to the characterization vector. The spatial relationship vector contains an angle of inclination between the first point cluster and the second point cluster. The invention is based on the objective of designing a method, a data processing device, a detection system, a driver assistance system, and a vehicle in which the recognition of objects from point cloud data can be improved. In particular, objects on the ground should be able to be better distinguished from reflections. Disclosure of the invention The object of the invention is achieved in the method by selecting at least one point of the point cloud as a focus point, for which at least one focus point at least one orientation value is determined, which specifies at least one relative orientation of a focus vector extending between the viewing location and the at least one focus point to at least one focus neighbor vector extending between the at least one focus point and at least one point of the point cloud selected as a neighbor point to the at least one focus point, and contour data describing a contour of at least one object is determined on the basis of at least two orientation values. According to the invention, at least one point of the point cloud is selected as the focus point. In this method, several points can also be selected successively or simultaneously as focus points. For at least one focus point, the relative orientation of a connecting vector between the focus point and at least one neighboring point is determined with respect to a connecting vector between the focus point and the viewing location. For simplicity, the connecting vector between the focus point and the at least one neighboring point is referred to as the focus-neighbor vector. Similarly, the connecting vector between the focus point and the viewing location is referred to as the focus vector. The relative orientation between these two connecting vectors is quantified with an orientation value. This method can also be used to determine the relative orientations of the connecting vectors for multiple neighboring points of a focus point and their respective orientation values. If similar, ideally identical, orientation values ​​are determined for three adjacent points—either a focal point and two of its neighboring points, or two focal points and one neighboring point each—it can be assumed that the three points correspond to reflection points on the same planar surface of an object. In this way, a planar surface of an object, such as a lost charge in the form of a box or the like, can be identified. If different, ideally significantly different, orientation values ​​are determined for three adjacent points—namely, either a focal point and two of its neighboring points, or two focal points and one neighboring point each—it can be assumed that the three points correspond to reflection points on different, mutually inclined surfaces. In this way, the surface of an object, particularly a lost cargo in the form of a box or the like, can be distinguished from the surface of the ground, especially the roadway, on which the object is located. According to the invention, contour data describing the contour of at least one object are determined based on at least two orientation values. These contours can be used to characterize the detected objects. The orientation values ​​allow points in the point cloud belonging to the same object to be grouped together. In this way, objects, particularly lost cargo, lying on the ground, especially on a roadway, can be distinguished from the ground itself. Advantageously, the contour data can be stored in at least one contour file. This makes the contour data easier to process further. Advantageously, contour data can be used for various applications of advanced driver assistance systems (ADAS). In particular, the contour data can be used for obstacle avoidance procedures, time-to-collision calculations, emergency braking assistants, and the like. For easier differentiation, individual points of the point cloud are referred to as focus points or neighboring points according to their respective function in the procedure. A focus point is the point in the point cloud that is considered first when determining an orientation value. The position of the focus point relative to the viewing location is described by the corresponding focus vector. A neighboring point is a point located in the vicinity of the selected focus point with respect to its respective coordinates. The position of a neighboring point relative to the focus point is described by the corresponding focus-neighbor vector. The viewing point is the location within the coordinate system from which the point cloud is viewed. The viewing point describes the position of the detection system used to capture the monitored area and generate the point cloud, relative to the coordinate system. A triangle is defined by the viewing location, the chosen focus point, and the corresponding neighboring point. The sides of the triangle are formed by the focus vector, the focus-neighbor vector, and a neighboring vector. The neighboring vector describes the position of the neighboring point relative to the viewing location. The at least one orientation value specifies the size of the angle of this triangle at the focus point, that is, the size of the angle enclosed between the focus vector and the focus-neighbor vector. The positions of the points in the point cloud and the observation point are specified by at least one spatial coordinate with respect to a coordinate system assigned to the point cloud. In this way, the positions can be described precisely. Advantageously, an orthogonal coordinate system can be assigned to the point cloud. In this way, the position of the points in the point cloud and the viewing location can be uniquely specified. Advantageously, the coordinate system can be a Cartesian coordinate system. This allows the positions of the points in the point cloud and the observation point to be specified by three coordinates, specifically x, y, and z. Alternatively, the coordinate system can be a spherical coordinate system. The focus vectors, the focus neighbor vectors, and the neighbor vectors can be determined from the position vectors of the focus points, the neighboring points, and the observation point. In this way, both the direction and magnitude of the respective vectors can be determined. Position vectors are known to be the vectors that specify a location in a coordinate system relative to the origin of the coordinate system. Advantageously, the point cloud can be a LiDAR point cloud. LiDAR point clouds can be generated through LiDAR measurements using LiDAR systems. The positions of the points in the LiDAR point cloud allow for a very precise description of the positions of reflective surfaces of objects within the monitored area, which were scanned by the LiDAR system's scanning beams. Alternatively, the point cloud can also be determined using a different measurement method and / or a different detection system, in particular a radar system or the like. The point cloud can also be generated by combining data obtained with different detection systems. Advantageously, the point cloud, in particular the LiDAR point cloud, was generated during the monitoring of at least one monitoring area with at least one detection system, in particular at least one LiDAR system. In this way, objects detected during monitoring in the monitoring area with the at least one detection system can be identified using the method according to the invention. Advantageously, the generation of the point cloud can be completed before the inventive method for recognizing objects from the point cloud data begins. In this way, all the point cloud data can be used in the method. Advantageously, the method can be used to detect objects, particularly lost cargo, on the ground, especially on a roadway. In this way, reflections of scanning signals from objects can be distinguished from reflections from the roadway. Thus, lost cargo on the ground, especially on the roadway, can be detected. Lost cargo refers to objects that have fallen from a preceding vehicle, especially a transport vehicle such as a truck or the like. The method according to the invention can distinguish objects from the ground, particularly the roadway, even in sparse point clouds. Due to its high efficiency, the method according to the invention can be performed with relatively few processor resources. Thus, even dense point clouds, especially those with more than 1 million points per image, can be examined in real time. The invention first identifies the common source of an object, particularly lost cargo, and distinguishes it from the points in the point cloud that originate from the ground, particularly the roadway. This can be done regardless of the type and condition of the ground, particularly the roadway. Thus, objects can be distinguished from the ground, particularly the roadway, even in complex ground topographies. This reduces the computational effort, and in particular the costs for corresponding data processing equipment. Advantageously, the detection system can operate according to a signal-time time method. Detection systems operating according to the signal-pulse-time time method can be designed and named as Time-of-Flight (TOF) systems, indirect Time-of-Flight (iTOF) systems, Light Detection and Ranging (LiDAR) systems, Laser Detection and Ranging (LaDAR) systems, Radio Detection and Ranging (Radar), or similar systems. The invention can be used in vehicles. A key functional characteristic of a vehicle is its ability to move. Vehicles can be motor vehicles. Advantageously, the invention can be used in land vehicles, in particular cars, trucks, buses, motorized or non-motorized two-wheelers such as motorcycles, e-bikes, bicycles, or the like, drones, mobile robots, mobile machinery, in particular construction or transport machinery such as cranes, excavators, or the like, aircraft, in particular drones, and / or (under)water vehicles, in particular (under)water drones. The invention can also be used in vehicles that can be operated autonomously or semi-autonomously. However, the invention is not limited to vehicles. It can also be used in stationary operation. Advantageously, the method according to the invention can be carried out within the framework of a perception system, particularly in conjunction with a driver assistance system. In this way, the perception of objects, especially objects detected by vehicle detection systems, particularly LiDAR systems, can be improved. This can increase the operational safety of the vehicles. In an advantageous embodiment of the method, at least one orientation value for the relative orientation of the respective focus vector and at least one focus neighbor vector to a neighboring point can be determined for multiple focus points. In this way, even extended object surfaces can be detected. Furthermore, a better resolution with respect to the orientation values ​​can be achieved. Overall, this improves the accuracy of object detection. Advantageously, several adjacent points can be selected as focus points, either sequentially or simultaneously. This allows for better detection of contiguous surfaces of objects. In a further advantageous embodiment of the method, at least two orientation values ​​for the relative orientations of the focus vector and corresponding focus-neighbor vectors to at least two neighboring points can be determined for at least one focus point. In this way, the relative orientations for at least one focus point in different directions in its vicinity can be determined more precisely. Thus, the spatial orientations of the surfaces of objects associated with the at least one focus point can be determined more accurately. In a further advantageous embodiment of the method, at least one orientation value for relative orientations can be determined for at least one focus point for at least one focus-neighbor vector of at least one neighboring point in the nearest neighborhood to the at least one focus point. In this way, jumps in orientation values ​​can be more accurately assigned to the corresponding locations on the surfaces of objects. Thus, transitions on the surfaces of objects, in particular transitions between the surface of lost charge and the ground surface, can be located more precisely. This allows for the determination of more accurate contour data, which enables a more precise description of the detected objects. In a further advantageous embodiment of the method, the orientation values ​​for relative orientations of focus-neighbor vectors to at least two neighboring points can be determined for at least one focus point, which lie in the same plane as the viewing location and the at least one focus point with respect to the coordinate system, and / or the orientation values ​​for relative orientations of the focus-neighbor vectors to at least two neighboring points can be determined for at least one focus point, which lie in different, in particular orthogonal, planes with respect to the viewing location with respect to the coordinate system. Advantageously, orientation values ​​for relative orientations can be determined for focus points, viewing locations, and neighboring points that lie in the same plane with respect to the coordinate system. In this way, transitions of surfaces of objects in the monitored area along a line can be detected. Advantageously, the plane containing the at least one focus point, the observation point, and the at least two neighboring points in the coordinate system can run parallel to a straight line, in particular a coordinate axis, which corresponds to a horizontal line in the monitored area. In this way, the horizontal path of a detected object can be identified using this method. Alternatively, the plane containing the at least one focus point, the observation point, and the at least two neighboring points in the coordinate system can advantageously run parallel to a straight line, in particular a coordinate axis, which corresponds to a vertical line in the monitored area. In this way, the vertical path of a detected object can be identified using the method. Advantageously, the horizontal and vertical orientations can be aligned with corresponding reference axes on the side of the detection system, in particular the LiDAR system, or, if applicable, on the vehicle on which the detection system is mounted. Advantageously, the horizontal direction can run parallel to a transverse axis of the vehicle. The vertical direction can run parallel to a vertical axis of the vehicle. Advantageously, orientation values ​​for relative orientations can be determined for focus points, viewing locations, and neighboring points that lie in different planes with respect to the coordinate system. In this way, the extent of objects in two spatial directions can be detected. Orienting the observation of neighboring points to a focal point along two orthogonal planes, particularly in the vertical and horizontal directions, improves the detection of lost cargo, especially in the form of boxes. At least one surface of a box lying on the ground, particularly the roadway, is oriented perpendicular to the ground. In a further advantageous embodiment of the method, contour data for the at least one object can be determined based on at least two orientation values ​​using at least one neural network, in particular at least one feedforward network. In this way, the contour data can be determined more efficiently, and in particular faster. The use of a feedforward network is especially efficient with regard to the runtime of the computation. In a further advantageous embodiment of the method, at least some of the alignment values ​​can be subjected to a scaling calculation, in particular a grayscale scaling, and the contour data for the at least one object can be determined from the scaled alignment values. In this way, the alignment values ​​can be processed more efficiently. For example, the scaled alignment values ​​can be fed into a neural network to determine the contour data. In a further advantageous embodiment of the method, at least one orientation value for at least one focus point can be determined as an angular quantity from the magnitude of the corresponding focus vector and / or the magnitude of a neighboring vector between the at least one neighboring point and the viewing location and / or the magnitude of the corresponding focus-neighbor vector, in particular using a cosine or sine function and / or at least one scalar product. In this way, the at least one orientation value can be determined efficiently and accurately. The angular magnitude indicates a measure of the relative orientation of the focus vector to the focus neighbor vector. The angular magnitude also includes the angle enclosed between the focus vector and the corresponding focus neighbor vector. The angle size can be calculated using the following formula: Here, Θ is the angular size of the angle, with 0≤ Θ≤π, between the focus vector and the focus neighbor vector, a is the magnitude of the focus vector, b is the magnitude of the neighbor vector, and c is the magnitude of the focus neighbor vector. Advantageously, at least one angle dimension can be subjected to a scaling calculation. The following formulas can be used for this purpose: Here, I is the scaled angle size, θ is the original angle size, and U is a predefined conversion value. The conversion value U can be predefined, in particular depending on the scaling increments, especially within the color channel. Advantageously, at least one angular size can be scaled to grayscale. Grayscale is particularly suitable for processing with neural networks. Advantageously, the conversion factor U = 255 can be specified. This corresponds to 256 gradations, or 8 bits, within the color channel. In a further advantageous embodiment of the method, a distance value determined from measurements with a detection system, in particular a LiDAR system, for the respective focus point can be used as the magnitude of the focus vector, and / or a distance value determined from measurements with a detection system, in particular a LiDAR system, for the respective neighboring point can be used as the magnitude of the neighboring vector. In this way, the magnitudes of the focus connection vectors and the neighboring connection vectors can be determined directly from the measurement without any further calculations or conversions. The neighbor vector is the connecting vector between the corresponding neighboring point and the point of observation. In a further advantageous embodiment of the method, the point cloud can be implemented as a file containing a multitude of data records, each characterizing at least one point of the point cloud, wherein each data record can contain at least one location data field with at least one location value, in particular at least one location coordinate. In this way, the point cloud data can be processed with a suitable data processing device. Advantageously, the points, and in particular the data records, can include at least one additional data field besides the at least one location data field. This allows for the description of additional information about objects in the point cloud. Advantageously, the points, and in particular the data records, can have at least one distance data field in which a distance value is specified. The distance value can specify the distance of a reflection point of an object corresponding to the point in the point cloud relative to a reference point of the detection system, in particular the LiDAR system, and / or a vehicle equipped with the detection system. Furthermore, the problem in the data processing device is solved by the fact that the data processing device has at least some means for carrying out the method according to the invention. The means for carrying out the method according to the invention is designed to select at least one point of a point cloud, in particular a LiDAR point cloud, as a focus point. Furthermore, the means for carrying out the method according to the invention for determining at least one orientation value for the at least one focus point are provided. This orientation value specifies at least one relative orientation of a focus vector to at least one focus neighbor vector. The focus vector extends between a viewing location and the at least one focus point. The viewing location is situated within the coordinate system and defines the location from which the point cloud is viewed. The at least one focus neighbor vector extends between the at least one focus point and at least one point in the point cloud selected as a neighbor to the at least one focus point. Furthermore, the means for carrying out the method according to the invention are designed for determining contour data which describe a contour of at least one object, based on at least two orientation values. At least some of the resources can be implemented through software. This allows existing hardware resources of the data processing system to be used. Alternatively or additionally, at least some of the resources can be implemented through hardware. Advantageously, all means for carrying out the method according to the invention can be implemented in the data processing unit. In this way, the method can be carried out centrally using the means of the data processing unit. Advantageously, the data processing device can be part of a detection system, in particular a LiDAR system or a radar system, and / or a driver assistance system and / or a vehicle. In this way, at least part of the means for carrying out the method according to the invention can be implemented using means of the detection system and / or the driver assistance system and / or the vehicle. Advantageously, at least part of the means for carrying out the method according to the invention can be implemented using cloud computing. In this way, at least part of the method can also be carried out outside of the data processing device, a detection system, a driver assistance system, and a vehicle. Furthermore, the object of the invention is achieved in the detection system by the fact that the detection system comprises at least some means for carrying out the method according to the invention. In this way, at least some part of the method according to the invention can be carried out using means of the detection system. Advantageously, all means for carrying out the method according to the invention can be implemented in the detection system. In this way, the method can be carried out centrally using means of the detection system. The detection system can be used to perform the measurements from which the point cloud can be generated, which is used in carrying out the method according to the invention. Furthermore, the object of the invention is achieved in the driver assistance system by the fact that the driver assistance system comprises at least some means for carrying out the method according to the invention. In this way, at least some part of the method according to the invention can be carried out using means of the driver assistance system. Advantageously, all means for carrying out the method according to the invention can be implemented with the driver assistance system. In this way, the method can be carried out centrally using means of the driver assistance system. The driver assistance system can be used for the autonomous or semi-autonomous control of functions, especially driving functions, of a vehicle. Advantageously, the driver assistance system can include or be connected to at least one detection system, in particular at least one LiDAR system and / or at least one radar system. In this way, information obtained by the at least one detection system, particularly regarding at least one monitoring area, can be used for autonomous or semi-autonomous control of the vehicle's functions. Furthermore, the object of the invention is achieved in the vehicle by the fact that the vehicle has at least some means for carrying out the method according to the invention. In this way, at least part of the process according to the invention can be carried out using means of the vehicle. Advantageously, all means for carrying out the method according to the invention can be implemented using the vehicle's own equipment. In this way, the method can be carried out centrally using the vehicle's equipment. If the vehicle has a driver assistance system, in particular a driver assistance system according to the invention, and / or at least one detection system, in particular a detection system according to the invention, and / or at least one data processing device, in particular a data processing device according to the invention, the means of the driver assistance system, the detection system, and the data processing device are also part of the means of the vehicle. The same applies to the means of the driver assistance system, the detection system, and the data processing device with respect to each other. Advantageously, the vehicle can have at least one detection system, in particular at least one LiDAR system and / or at least one radar system. In this way, at least one monitoring area, particularly in the vicinity of the vehicle, can be monitored for objects. The measurements from which the point cloud used in carrying out the method according to the invention can be generated can be performed with the at least one detection system. Advantageously, the vehicle can have at least one driver assistance system. This system allows the vehicle's functions, particularly driving functions, to be controlled autonomously or semi-autonomously. If the vehicle has a driver assistance system, in particular a driver assistance system according to the invention, and / or at least one detection system, in particular a detection system according to the invention, and / or at least one data processing device, in particular a data processing device according to the invention, the means of the driver assistance system, the detection system, and the data processing device are also part of the means of the vehicle. The same applies to the means of the driver assistance system, the detection system, and the data processing device with respect to each other. Furthermore, the features and advantages identified in connection with the inventive method, the inventive data processing device, the inventive detection system, the inventive driver assistance system, and the inventive vehicle and their respective advantageous embodiments apply to each other accordingly, and vice versa. The individual features and advantages can, of course, be combined with one another, potentially resulting in further advantageous effects that go beyond the sum of the individual effects. Brief description of the drawings Further advantages, features, and details of the invention will become apparent from the following description, in which exemplary embodiments of the invention are explained in more detail with reference to the drawings. The person skilled in the art will expediently consider the features disclosed in the drawing, the description, and the claims individually and combine them into meaningful further combinations. Figure 1 schematically shows a vehicle with a driver assistance system comprising a LiDAR system in a driving situation in which an object lies on the road in front of the vehicle in the direction of travel; Figure 2 shows a functional diagram of the vehicle's driver assistance system from Figure 1; Figure 3 shows a point cloud generated of the object by the vehicle's LiDAR system; Figure 4 shows a file describing a point cloud generated from Figure 1 by the vehicle's LiDAR system; Figure 5 shows...5 a method for detecting objects from point cloud data generated by the vehicle's LiDAR system in Fig. 1; Fig. 6 a passive image of a situation in which three objects are located at different distances in front of the vehicle in Fig. 1 on the roadway; Fig. 7 a visualization of an inclination angle file generated by the object detection method from Fig. 5 of the situation depicted in the image from Fig. 6; Fig. 8 a visualization of a contour file containing the contour data for the objects in the situation from Fig. 6. In the figures, identical components are labelled with the same reference symbols. embodiment(s) of the invention Figure 1 shows a vehicle 10 located on a roadway 12. In the direction of travel, in front of the vehicle 10, is an object 14, for example, lost cargo from a preceding vehicle (not shown). The object 14 is lying on the roadway 12. The vehicle 10 has a driver assistance system 16. With the driver assistance system 16, functions of the vehicle 10, such as driving functions, can be controlled autonomously or semi-autonomously. The driver assistance system 16 comprises a detection system in the form of a LiDAR system 18 and a central control unit 20. The LiDAR system 18 is connected to the central control unit 20 for control and data transmission. Fig. 2 shows a functional diagram of the driver assistance system 16 with the LiDAR system 18. The LiDAR system 18 is located in the front part of the vehicle 10, for example, in a bumper. The LiDAR system 18 can monitor a surveillance area 22 in the direction of travel in front of the vehicle 10. For example, the LiDAR system 18 can detect objects 14 within the surveillance area 22. The LiDAR system 18 can also be located elsewhere on the vehicle 10, even in a different orientation. In addition to or as an alternative to the LiDAR system 18, another detection system, such as a radar system or the like, can also be provided. The LiDAR system 18 comprises a transmitter 24, a receiver 26 and a control and evaluation unit 28. The transmitter 24, the receiver 26 and the control and evaluation unit 28 are interconnected for control and data transmission. The transmitter 24 can send scanning light signals 30, for example in the form of laser pulses, into the monitoring area 22. Scanning light signals 30 that strike a surface in the monitoring area 22, for example a surface of the object 14 or a surface of the roadway 12, are reflected. Scanning light signals 30 reflected towards the LiDAR system 18 can be received by the receiver 26 as echo light signals 32. The echo light signals 32 can be converted by the receiver 26 into electrical signals, which can be supplied to the control and evaluation unit 28. The LiDAR system 18 can, for example, be operated using a time-of-flight method. With the control and evaluation unit 28, distances, velocities, and directions of objects 14, from which the scanning light signals 30 are reflected, relative to a reference point of the LiDAR system 18 can be determined based on the scanning light signals 30 and the echo light signals 32. The reference point of the LiDAR system 18 can also be considered the reference point of the vehicle 10. The control and evaluation unit 28 includes a conversion device, for example in the form of a decoder 34, and a data processing unit 36. The decoder 34 can convert the electrical signals associated with the scanning light signals 30 and the echo light signals 32, for example electrical signals coming from the receiving unit 26, into a format that can be processed by the data processing unit 36. The data processing unit 36 ​​can then generate a point cloud 38 from the converted data. Figure 3 shows an example of a point cloud 38 for the object 14 shown in Figure 1 on the roadway 12. For ease of orientation, the outline of the object 14 and the course of the roadway 12 are indicated by dashed lines. The outline of the object 14 is not part of the point cloud 38. Since the point cloud 38 was determined by LiDAR measurements with the LiDAR system 18, it can also be referred to as a LiDAR point cloud. The point cloud 38 comprises a multitude of points 40. Each point 40 originates from an echo light signal 32 that was reflected from a reflection point on the surface of the object 14 or the roadway 12. Thus, each point 40 can be assigned to a reflection point of an object 14 or the roadway 12. In Fig. 3, one of the points 40 is labelled with the index F. Two of the points 40 are labelled with the index N. Another point 40 is labelled with the index R. The indices F, N, and R serve, as will be explained in more detail below, to facilitate the differentiation of these specific points 40. Points 40F and 40N belong to the surface of object 14. Points 40R belong to reflections on the roadway 12. The point cloud 38 is stored as file 48 in the control and evaluation unit 28. Figure 4 shows an example of the structure of file 48. Each of the points 40 is characterized by a data record 50. Each data record 50 comprises, by way of example, three location data fields 52O, one intensity data field 52I, and one distance data field 52D. The location data fields 52O contain the coordinates of the respective point 40 with respect to a coordinate system 42. The coordinate system 42 is assigned to the point cloud 38. A Cartesian xyz coordinate system 42 is provided as an example, which is indicated in Fig. 3. The coordinates of the location data fields 52O are the x-coordinate, y-coordinate, and z-coordinate of the respective point 40. The x-axis of coordinate system 42 runs parallel to the transverse axis of vehicle 10. In the normal operating orientation of vehicle 10, the x-axis is horizontal with respect to the roadway 12. The y-axis runs parallel to the longitudinal axis of the vehicle. The z-axis runs parallel to the vertical axis of the vehicle. In the normal operating orientation, the z-axis is vertical with respect to the roadway 12. The origin 44 of coordinate system 42 is located, for example, in the upper left front corner of the image of the monitored area 22, as viewed from the LiDAR system 18. The origin 44 can also be located elsewhere, for example, at a viewing point 46 of the LiDAR system 18. The observation point 46 is the location in the coordinate system 42 that corresponds to the reference point of the LiDAR system 18. From this reference point, the monitoring area 22 is observed using the LiDAR system 18. Accordingly, the point cloud 38 is observed from observation point 46. Observation point 46 is not part of the point cloud 38. Observation point 46 is described by a separate data set containing three location data fields for the x, y, and z coordinates of the coordinate system 42. The intensity data fields 52Ider datasets 50 each contain the intensities of the received echo light signals 32, which are assigned to the respective point 40. Alternatively, other quantities, such as energy, luminous intensity, or the like, can also be specified there, which characterize the strength of the received light signals 32. The distance data fields 52D contain the distance values ​​corresponding to the distances between the reflection points of the object 14 or the roadway 12, of which the echo signals 32 belonging to the respective points 40, and the reference point of the LiDAR system 18. The distances can be determined, for example, directly by time-of-flight measurements of the scanning light signals 30 and corresponding echo light signals 32. A particular challenge for LiDAR perception algorithms is the detection of objects 14 in the form of lost charge, even at large distances. Therefore, it is essential to distinguish the objects 14 in the point cloud 38 from reflections originating from the ground, for example, the roadway 12. In the following, a method 54 for detecting objects 14, for example lost cargo, from data of a point cloud 38 in conjunction with a flowchart from Fig. 5 is explained in more detail. In a preliminary selection step 56 of the procedure 54, areas of the point cloud 38 are determined which could correspond to potential fixed objects 14, such as lost cargo, on the roadway 12. Fig. 3 shows an example of the selection of the point cloud 38 for the object 14 from Fig. 1 in the form of a box. In a horizontal angle calculation step 58, respective horizontal angle values ​​60 are calculated for each of the points 40 based on the location values ​​of the location data fields 52 of the points 40. The respective horizontal angle value 60 quantifies a horizontal inclination angle 62. In Fig. 3, one of a total of two horizontal inclination angles 62 for one of the points 40 is indicated as an example; for better differentiation, this point is designated as focus point 40F. The horizontal tilt angle 62 is enclosed between a focus vector 64 and a focus neighbor vector 68. The focus vector 64 extends from the viewing location 46 to the focus point 40F. The focus neighbor vector 68 extends between the focus point 40F and the nearest neighbor point 40N in a direction parallel to the x-axis, i.e., in the horizontal direction. The horizontal angle values ​​60 are alignment values ​​for a relative alignment of the respective focus vector 64 and the respective focus neighbor vector 68. Every point 40 – except for the points 40 at the edge of the point cloud 38 – has two nearest neighboring points 40N in the horizontal direction. Therefore, two horizontal angles 60 can also be determined for the corresponding horizontal tilt angles 62. When determining the two horizontal angles 60 for a focus point 40F, the corresponding neighboring points 40N, the focus point 40F, and the viewing location 46 lie in a plane. The magnitude of the focus vector 64 is the distance value D of data set 50 of focus point 40F. Alternatively, the magnitude of the focus vector 64 can be calculated as the difference between a position vector of the viewing location 46 and the position vector of focus point 40F. The position vectors of points 40, namely focus points 40F and neighboring points 40N and of the observation location 46, are each specified by the coordinates, namely x, y and z, in the position data fields 52 of the respective points 40 and of the observation location 46. Similarly, the distance value D of data set 50 of the neighboring point 40N is used as the magnitude of the neighboring vector 66. Alternatively, the magnitude of the neighboring vector 66 can also be calculated here as the difference between the position vector of the observation point 46 and the position vector of the neighboring point 40N. The magnitude of the focus-neighbor vector 68 is calculated as the difference between the position vector of the focus point 40F and the position vector of the neighbor point 40N. The calculation of the horizontal angle size 60 is performed according to the following formula: The horizontal angle size 60 is calculated according to the following formula: where Θ is the horizontal angle size 60 of the horizontal inclination angle 62, with 0≤ Θ≤π, between the focus vector 64 and the focus neighbor vector 68, a is the magnitude of the focus vector 64, b is the magnitude of the neighbor vector 66 and c is the magnitude of the focus neighbor vector 68. In a vertical angle calculation step 70, the corresponding vertical angle quantities 72 are calculated for each of the points 40. The vertical angle quantities 72 quantify the vertical inclination angles 74 for each of the points 40 considered as focus point 40F relative to the nearest neighboring points 40N in the vertical direction, for example, in the direction of the z-axis. In Fig. 2, the vertical inclination angle 74 for the focus point 40F shown there is indicated by a dashed line. In a vertical angle calculation step 70, respective vertical angle values ​​72 are calculated for each of the points 40 based on the position values ​​of the location data fields 52 of the points 40. The respective vertical angle value 72 quantifies a vertical tilt angle 74. In Fig. 3, one of a total of two vertical tilt angles 74 for point 40, which was already selected as the focus point 40F in the calculation of the horizontal tilt angles 62 described above, is indicated by a dashed line. The vertical tilt angle 74 is enclosed between the focus vector 64 and the focus neighbor vector 68 to the nearest neighbor point 40N in a direction parallel to the z-axis, i.e., in the vertical direction. The calculation of the vertical angle values ​​72 is analogous to the calculation of the horizontal angle values ​​60. The vertical angle values ​​72 are alignment values ​​for a relative alignment of the respective focus vector 64 and the respective focus neighbor vector 68. Every point 40 – except for the points 40 at the edge of the point cloud 38 – has two nearest neighboring points 40N in the vertical direction. Therefore, two vertical angles 72 can also be determined for the corresponding vertical inclination angles 74. When determining the two vertical angles 72 for a focus point 40F, the corresponding neighboring points 40N, the focus point 40F, and the viewing location 46 lie in a plane. When determining the horizontal angle sizes 60 and the vertical angle sizes 72 for a focus point 40, the vertical neighboring points 40N and the horizontal neighboring points 40N fly in different, mutually perpendicular planes. Subsequently, in a combination step 76, the horizontal angle sizes 60 and the vertical angle sizes 72 of the points 40 are combined in an inclination angle file 78. Optionally, the horizontal angle values ​​60 and the vertical angle values ​​72 are subjected to grayscale scaling as part of the combination. The following calculation formulas are used for this purpose: Here, I is the scaled angle size, θ is the original angle size (horizontal angle size 60° or vertical angle size 72°), and U is a predefined conversion value. The conversion value U is predetermined depending on the scaling increments, for example, within the color channel. As an example, the conversion factor U = 255 is predefined. This corresponds to 256 increments, or 8 bits, within the color channel. A visualization of an example tilt angle file 78 is shown in Fig. 7, which is explained in more detail below. In contour determination step 80, the data from the inclination angle file 78 are fed to a contour determination system 82. The contour determination system 82 is based on artificial intelligence. The contour determination system 82 is designed as a neural network, for example, as a feedforward network. The contour determination system 82 is part of the data processing unit 36. At the output of the contour detection system 82, contour data 84 is output, which describes the contours 86 of the detected object 14. The contour data 84 is stored, for example, in a contour file. A visualization of the contour file with the contours 86 is shown in Fig. 8, described below. Using the contour data 84, the points 40 belonging to objects 14 can be assigned to clusters. The points 40 of objects 14 assigned to the clusters can thus be distinguished from the ground, namely the roadway 12. The contour file containing contour data 84 is provided to the central control unit 20 of the driver assistance system 16. Based on the contour file, various applications of the driver assistance system, such as obstacle avoidance procedures, procedures for determining the time until collision, an emergency braking assistant, or the like, can be implemented. Figure 6 shows a passive grayscale image of a situation in front of the vehicle 10. "Passive" means that the image was captured without additional illumination, such as scanning light signals 30. In this image, three objects 14, for example lost cargo, are located at different distances in front of the vehicle 10 in the direction of travel. For ease of orientation, the images of the objects 14 in Figure 6 are labeled with the same reference symbols as the objects 14. Fig. 7 shows a visualization of an inclination angle file 78, which was determined from a point cloud 38 using the method 54 described above. This point cloud was acquired during a LiDAR measurement with the LiDAR system 18 in the situation shown in Fig. 6. The x-coordinates of the coordinate system 42 are plotted horizontally and the z-coordinates vertically. Points 40 with similar, ideally identical, horizontal angles 60 and similar, ideally identical, vertical angles 72 are marked with the same pattern in Fig. 7. If several points 40 in the vicinity have similar horizontal angles 60 and vertical angles 72, they are recognized as belonging to the same surface of an object 14. In Fig. 7, the areas with the points 40 corresponding to the three objects 14 from the situation shown in Fig. 6 are shown as examples.The situations shown in section 6 are each marked with reference 14 for easier differentiation. Fig. 8 shows the visualization of a contour file with the contour data 84 for the objects 14 from the situation shown in Fig. 6. For ease of orientation, the contours 86 for the three objects 14 are shown and labeled. In the representation in Fig. 8, the x-coordinate is plotted horizontally. The distance values ​​D for the respective points 40 are plotted vertically. The contours 86 for the two front objects 14 have a distance value D of approximately 50 m to the LiDAR system 18. The rear object 14 has a distance value D of approximately 100 m. In an alternative embodiment of the method 54, not shown, the vertical angle calculation step 70 is omitted. Similarly, in the combination step 76, the combination of the horizontal angle parameters 60 with the vertical angle parameters 72 is omitted. Only the grayscale scaling is performed. Only the horizontal angle parameters 60 are used to determine the contour data 84. QUOTES INCLUDED IN THE DESCRIPTION This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited patent literature US 2021 / 0272305 A1

[0006]

Claims

Method (54) for detecting objects (14) from data of a point cloud (38), in particular a LiDAR point cloud (38), wherein each point (40, 40F, 40N, 40R) of the point cloud (38) has at least one spatial coordinate (x, y, z) which indicates the position of the point (40, 40F, 40N, 40R) in a coordinate system (42) assigned to the point cloud (38), wherein in the method (54) at least one orientation of a vector (64) between a viewing location (46) within the coordinate system (42) and the at least one point (40F) is determined for at least one point (40F) of the point cloud (38), characterized in that at least one point (40F) of the point cloud (38) is selected as a focus point (40F), for which at least one focus point (40F) at least one orientation value (60, 72) is determined. which at least one relative orientation of a focus vector (64) extending between the viewing location (46) and the at least one focus point (40F),to at least one focus neighbor vector (68) extending between the at least one focus point (40F) and at least one point (40) of the point cloud (38) selected as a neighbor point (40N) to the at least one focus point (40F), is determined on the basis of at least two orientation values ​​(60, 72) contour data (84) which describe a contour (86) of at least one object (14). Method according to claim 1, characterized in that for several focus points (40F) at least one orientation value (60, 72) is determined for a relative orientation of the respective focus vector (64) and at least one focus neighbor vector (68) to a neighbor point (40N). Method according to claim 1 or 2, characterized in that for at least one focus point (40F) at least two orientation values ​​(60, 72) for relative orientations of the focus vector (64) and corresponding focus neighbor vectors (68) to at least two neighbor points (40N) are determined. Method according to one of the preceding claims, characterized in that for at least one focus point (40F) at least one orientation value (60, 72) for relative orientations for at least one focus neighbor vector (68) of at least one neighbor point (40N) in the nearest neighborhood to the at least one focus point (40F) is determined. Method according to one of the preceding claims, characterized in that for at least one focus point (40F) the orientation values ​​(60, 72) for relative orientations for focus-neighbor vectors (68) to at least two neighbor points (40N) are determined, which lie in the same plane with the viewing location (46) and the at least one focus point (40F) with respect to the coordinate system (42), and / or for at least one focus point (40F) the orientation values ​​(60, 72) for relative orientations of the focus-neighbor vectors (68) to at least two neighbor points (40N) are determined, which lie in different, in particular orthogonal, planes with respect to the viewing location (46) with respect to the coordinate system (42). Method according to one of the preceding claims, characterized in that contour data (84) for the at least one object (14) are determined on the basis of at least two orientation values ​​(60, 72) using at least one neural network (82), in particular at least one feedforward network. Method according to one of the preceding claims, characterized in that at least a part of the alignment values ​​(60, 72) is subjected to at least one scaling calculation, in particular a grayscale scaling, and the contour data (84) for the at least one object (14) are determined from the scaled alignment values ​​(60, 72). Method according to one of the preceding claims, characterized in that at least one alignment value (60, 72) for at least one focus point (40F) is determined as an angular quantity from the magnitude of the corresponding focus vector (64) and / or the magnitude of a neighbor vector (66) between the at least one neighbor point (40N) and the viewing location (46) and / or the magnitude of the corresponding focus neighbor vector (68), in particular using a cosine or sine function and / or at least one scalar product. Method according to one of the preceding claims, characterized in that a distance value (D) determined from measurements with a detection system (18), in particular a LiDAR system (18), for the respective focus point (40F) is used as the magnitude of the focus vector (64), and / or a distance value (D) determined from measurements with a detection system (18), in particular a LiDAR system (18), for the respective neighboring point (40N) is used as the magnitude of the neighboring vector (66). Method according to one of the preceding claims, characterized in that the point cloud (38) is realized as a file (48) which has a plurality of data records (50) which each characterize at least one point (40, 40F, 40N, 40R) of the point cloud (38), wherein each data record (50) has at least one location data field (50O) with at least one location value, in particular at least one location coordinate (x, y, z). Data processing device (36), in particular data processing device (36) for a LiDAR system (18), in particular for a vehicle (10), with at least part of means (28, 36, 82) for carrying out a method (54) for detecting objects (14) from data of point clouds (38), in particular of LiDAR point clouds (38), characterized in that the data processing device (36) has at least part of means (28, 36, 82) for carrying out the method (54) according to one of claims 1 to 10. Detection system (18), in particular LiDAR system (18), in particular detection system (18) for a vehicle (10), with at least one transmitting device (24) for transmitting scanning signals (30) into a monitoring area (22), at least one receiving device (26) for receiving scanning signals (30) reflected in the monitoring area (22) as echo signals (32) and at least one data processing device (36), in particular a data processing device (36) according to claim 11, with at least one part of means (28, 36, 82) for carrying out a method (54) for detecting objects (14) from data of point clouds (38), in particular of LiDAR point clouds (38), characterized in that the detection system (18) has at least one part of means (28, 36, 82) for carrying out the method (54) according to one of claims 1 to 10. Driver assistance system (16) with at least part of means (28, 36, 82) for carrying out a method (54) for detecting objects (14) from point cloud data (38), in particular LiDAR point clouds (38), characterized in that the driver assistance system (16) has at least part of means (28, 36, 82) for carrying out the method (54) according to one of claims 1 to 10. Vehicle (10) with at least one detection system (18), in particular at least one LiDAR system (18), for monitoring a monitoring area (22) in the vicinity of the vehicle (10) and with at least part of means (28, 36, 82) for carrying out a method (54) for detecting objects (14) from point cloud data (38), in particular LiDAR point clouds (38) which are generated with the at least one detection system (18), characterized in that the vehicle (10) has at least part of means (28, 36, 82) for carrying out the method (54) according to one of claims 1 to 10.

Images