METHOD FOR TRACKING A DISTANT TARGET VEHICLE IN THE SURVIVAL AREA OF A MOTOR VEHICLE BY MEANS OF A COLLISION DETECTION DEVICE

Description

[0001] The invention relates to a method for tracking a remote target vehicle in an area surrounding a motor vehicle by means of a collision detection device of the motor vehicle for detecting an impending collision with the target vehicle according to the preamble of claim 1. The invention further relates to a collision detection device for a motor vehicle for detecting an impending collision with a remote target vehicle.

[0002] Modern motor vehicles are increasingly using sensors to detect the surroundings, particularly the traffic situation and other road users. Such environmental sensors are typically part of a collision detection system for detecting an impending collision. Radar sensors, lidar sensors, camera sensors, or similar sensors are commonly used as environmental sensors. Environmental detection using a radar sensor, for example, is based on the emission of focused electromagnetic waves and their reflection, particularly by structures along the road or other vehicles.

[0003] It is also known from the prior art to activate a passive safety system of a motor vehicle, for example an airbag or a belt tensioner, in the event of a collision with a collision object, for example a vehicle. The safety system is usually triggered shortly after the time of impact, i.e. only after the collision has already occurred. However, as explained above, collision detection devices are already known which, by using appropriate sensors and evaluating their signals, detect a point in time at which an impact is unavoidable. This point in time is in the so-called pre-crash phase and therefore before the actual time of the collision. Using this information, the safety system is activated shortly before the collision in order to be able to protect the occupants of a motor vehicle even better from injury.

[0004] With regard to the activation of a passive safety system, particularly in the pre-crash phase, it is advantageous if not only collision-relevant data from the driver's own vehicle is known, but also collision-relevant data from a target vehicle with which a collision is imminent. Such collision-relevant data of a target vehicle includes, for example, speed, acceleration, yaw rate, and wheel steering angle. Since this collision-relevant data can change significantly, especially immediately before a collision, for example due to a sudden steering and / or braking maneuver or a skidding situation, it is essential to track the target vehicle sufficiently reliably, even in close range to the vehicle, particularly within a distance of less than approximately 2 meters, in order to determine the data as accurately as possible.

[0005] US 2016 / 291145 A1 describes a generic method for tracking a remote vehicle traveling in a lateral position relative to a host vehicle, wherein target data is provided by two radar sensors and fed to an object detection fusion system of the host vehicle. The wheels on the remote vehicle are identified as a grouping of radar points that have substantially the same location but substantially different Doppler range change values. For this purpose, radar points are grouped into clusters based on positional proximity, and for each cluster, the Doppler value of all radar points in the corresponding cluster is evaluated. Those clusters that have significantly different Doppler range change values ​​are identified as a wheel.If both wheels on the near side of the distant vehicle can be identified, a fusion calculation is performed using the wheel positions measured by both radar sensors, resulting in an accurate estimate of the distant vehicle's position, orientation, and speed. The distant vehicle's position, orientation, and speed are used to trigger warnings or evasive maneuvers in a lateral collision avoidance system.

[0006] EP 3 285 230 A1 describes a method for determining a motion state of an object, in particular a vehicle, outside a host vehicle. Based on acquired data from a radar sensor and a camera and / or a lidar sensor of the host vehicle, several feature points on the object and their respective positions and velocities are determined. Based on the determined positions and velocities, a reference position, a reference velocity, and a reference angular velocity, which represent a motion state for the object, are determined.

[0007] DE 10 2012 021 461 A1 describes a method for classifying vehicles, wherein at least one wheel of a vehicle is detected by means of at least one radar sensor. The radar sensor generates a superposed overall signal from several scattering centers of the wheel, which is formed from a characteristic pattern of radar radiation reflected by the scattering centers during rotation of the wheel. A wheel signature dependent on the scattering centers is then determined from the overall signal. A wheel size is derived from the signature, which is used to determine a vehicle size, vehicle mass, and / or vehicle category.

[0008] EP 2 538 239 A1 describes a method for detecting vehicle wheels using a stationary radar sensor or lidar sensor. An electromagnetic measuring beam with a known temporal profile of its frequency is emitted from the side of the roadway to an area above the roadway. A measuring beam reflected from a passing vehicle is received, and all of its frequencies are recorded and evaluated over time relative to the known profile. A wheel is detected if a frequency spread occurring during the vehicle's passage exceeds a specified spread value in the recorded profile.

[0009] Based on this, it is the object of the invention to provide a method for tracking a distant target vehicle in an area surrounding a motor vehicle and a corresponding collision detection device, by means of which a sufficiently accurate tracking of the target vehicle in a close range and, based thereon, in particular a sufficiently accurate determination of collision-relevant data is possible.

[0010] The above object is achieved by the entire teaching of claim 1 and the independent claim 17. Advantageous embodiments and further developments of the invention are set forth in the subclaims and the following description.

[0011] In the method according to the invention for tracking a remote target vehicle in a surrounding area of ​​a motor vehicle by means of a collision detection device of the motor vehicle to detect an impending collision with the target vehicle by fusing radar sensor data from a first environmental sensor designed as a radar sensor with sensor data from a second environmental sensor of the collision detection device, the radar sensor and the second environmental sensor each have detection areas that overlap such that at least a partial area of ​​the target vehicle is detected by both environmental sensors. The method comprises the following steps: Providing radar sensor data and sensor data of the target vehicle by means of the radar sensor and the second environmental sensor, wherein radar reflection points are derived from the radar sensor data, which determine a position determined by a distance and an azimuth angle and a Doppler velocity of points on the target vehicle, identifying a wheel of the target vehicle from the radar reflection points, providing first wheel detection data based on the wheel identified from the radar reflection points, identifying a wheel of the target vehicle from the sensor data of the second environmental sensor and providing second wheel detection data based on the wheel identified from the sensor data, merging the first wheel detection data and the second wheel detection data, determining a parameter of the target vehicle based on the merged first and second wheel detection data.

[0012] According to the invention, identifying a wheel of the target vehicle comprises placing a uniformly large two-dimensional window with an extension in the distance dimension and the azimuth angle dimension around each radar reflection point, determining a sum of the variances of the Doppler velocities of all radar reflection points contained in the window and assigning this sum to the corresponding radar reflection point, and determining a radar reflection point as a point of a wheel whose assigned sum is greater than a predetermined threshold value.

[0013] The invention is initially based on the consideration that the accuracy of tracking a distant target vehicle can be increased if the tracking is not based solely on the sensor data of a single environmental sensor. Furthermore, the invention is based on the consideration that a wheel of a target vehicle has a particular movement characteristic which, due to the micro-Doppler effect, is reflected in the radar sensor data acquired by a radar sensor, and whereby a wheel and, based on this, a parameter of the target vehicle can be reliably identified. This utilizes the fact that the movement of a wheel deviates from the movement of the body of the target vehicle or rotates in addition to the movement of the body of the target vehicle, generating an analyzable Doppler modulation around the main Doppler shift.

[0014] Therefore, the invention provides for the use of two environmental sensors to identify a wheel of the target vehicle and a fusion of the corresponding sensor data, wherein a first environmental sensor is designed as a radar sensor. Particularly reliable identification of a wheel using a radar sensor is achieved by placing a uniformly sized two-dimensional window with an extension in the distance dimension and the azimuth angle dimension around each radar reflection point and determining a sum of the variances of the Doppler velocities of all radar reflection points contained in the window and assigning this sum to the corresponding radar reflection point, wherein a radar reflection point is determined as a point of a wheel whose assigned sum is greater than a predetermined threshold value. The deviation of the velocity components caused by a (rotating) wheel is thus determined and evaluated.The aim here is to achieve a kind of normalization through the uniformly large two-dimensional window placed over each radar reflection point, and to assign a further parameter to each radar reflection point with the sum of the variances of the Doppler velocities of all radar reflection points contained in the window, which enables a reliable evaluation.

[0015] The embodiment according to the invention has the advantage that it provides a method by means of which a sufficiently accurate tracking of the target vehicle in a close range and, based thereon, in particular a sufficiently accurate determination of collision-relevant data is possible.

[0016] In the context of this invention, determining can be understood, for example, as calculating. Such calculating can be performed, for example, in a microcontroller, a microprocessor, or another programmable or hard-wired unit. It should also be understood that the remote target vehicle being tracked is typically not the vehicle executing the method.

[0017] Determining or calculating can also be understood in particular as meaning that not only a date or data is determined as such, but rather that it is determined or calculated as a probability distribution or in another way with reference to the probabilities of certain values ​​occurring. For example, corresponding values ​​can also be estimated.

[0018] Preferably, all wheels of the target vehicle are identified for which this is possible, particularly depending on the detection ranges of the first environmental sensor designed as a radar sensor and the second environmental sensor. For each identified wheel, first wheel detection data and second wheel detection data are then determined or provided accordingly.

[0019] The two-dimensional window can in particular be a so-called sliding window known to those skilled in the art from radar technology, which passes through the radar reflection points one after the other or is placed around each radar reflection point.

[0020] In an advantageous embodiment, deriving the radar reflection points from the radar sensor data comprises the following steps: Determining radar reflection points by means of Fourier transformation from the radar sensor data, separating the radar reflection points from the noise of the radar sensor data using a CFAR filter, and identifying the radar reflection points reflected by the target vehicle by means of a clustering method, preferably a DBSCAN method.

[0021] In this case, a three-dimensional Fourier spectrum of radar reflection points with distance, azimuth angle, and Doppler velocity is advantageously created using Fourier transformation. Separating the radar reflection points from the noise of the radar sensor data using a CFAR filter (constant false alarm rate filter) serves to identify the radar reflection points reflected by an object, and in particular by the target vehicle, and to distinguish them from the signal noise contained in the radar sensor data. Identifying the radar reflection points reflected by the target vehicle using a clustering method is based on viewing the individual radar reflection points as a unified whole, which represents the target vehicle. A clustering algorithm designed as DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is preferably used as the clustering method.DBSCAN is a spatial cluster analysis based on a density-based algorithm that can detect one or more clusters.

[0022] In a further advantageous embodiment, the determination of the sum of the variances of the Doppler velocities of all radar reflection points contained in a window comprises the following steps: Determining an arithmetic mean of the Doppler velocities of all radar reflection points contained in the window, determining a variance of the Doppler velocity of a radar reflection point as the difference between the Doppler velocity of the radar reflection point and the mean for all radar reflection points contained in the window, and forming the sum of the determined variances of the Doppler velocities of all radar reflection points contained in the window.

[0023] In this way, the sum of the variances of the Doppler velocities of all radar reflection points contained in a window can be determined particularly accurately and reliably.

[0024] In a further advantageous embodiment, identifying a wheel of the target vehicle from the radar reflection points further comprises the following steps: Assigning the radar reflection points determined as a point of a wheel to a wheel cluster by a clustering method, preferably a DBSCAN method, and determining the wheel cluster as a wheel of the target vehicle.

[0025] For example, if the target vehicle is a two-track, two-axle vehicle, a maximum of four wheel clusters can be generated, depending on the detection range of the radar sensor.

[0026] In a further advantageous embodiment, the provision of first wheel detection data comprises the following steps: Calculating the position data of a wheel center of gravity based on the radar reflection points assigned to a wheel cluster, and providing the position data of the wheel center of gravity as initial wheel detection data.

[0027] The wheel's center of gravity or the position data of the wheel's center of gravity represent the wheel's position. The position data are represented, in particular, by an x-value and a y-value in a reference coordinate system of the motor vehicle. The position data are preferably determined by averaging the corresponding x-values ​​and y-values ​​of the radar reflection points assigned to the (common) wheel cluster.

[0028] In a further advantageous embodiment, the provision of first wheel detection data further comprises the following steps: Determining an azimuth angle value of the radar reflection point positionally closest to the wheel center of gravity, grouping the radar reflection points that have the determined azimuth angle value as an azimuth angle value into a group, determining the radar reflection point of the group that has a relatively highest intensity value within the group and that has not been determined as a point of a wheel, providing the Doppler velocity of the determined radar reflection point as the velocity of the target vehicle at the position of the wheel center of gravity as first wheel detection data.

[0029] In this way, the speed of the target vehicle at the position of the wheel's center of gravity can be determined and provided with relative accuracy. Since there are many points on the target vehicle's body with a higher reflectivity than the wheels, a search is essentially performed for the measured maximum intensity among all radar reflection points that have the same or approximately the same azimuth angle as the wheel or wheel's center of gravity in question and that have not been identified as a single point on a wheel. This ensures, as far as possible, that the radar reflection point determined in this way actually corresponds not to a point on the wheel, but specifically to a point on the body of the target vehicle.

[0030] On the basis of the Doppler velocity thus determined and provided as the speed of the target vehicle at the position of the wheel center of gravity, a rotation rate of the target vehicle can also advantageously be determined.

[0031] In a further advantageous embodiment, the merging of the first wheel detection data and the second wheel detection data comprises the following steps: Assigning the first wheel detection data and the second wheel detection data to a target vehicle axle and a target vehicle side, respectively, merging the first wheel detection data and second wheel detection data assigned to a common target vehicle axle and target vehicle side based on a covariance intersection algorithm.

[0032] The use of a covariance intersection algorithm in the step of merging the first and second wheel detection data associated with a common target vehicle axle and target vehicle side ensures a convergent fusion of the first wheel detection data with the second wheel detection data.

[0033] To assign the first wheel detection data and the second wheel detection data, wheel gates are advantageously generated, to which the first wheel detection data and the second wheel detection data are each assigned, in particular based on position data of the first wheel detection data and the second wheel detection data. For a two-track, two-axle vehicle as the target vehicle, four wheel gates are generated, namely one wheel gate each for the front-right, front-left, rear-right, and rear-left, with the first wheel detection data and the second wheel detection data each being assigned to the corresponding wheel gate.

[0034] Advantageously, in the step of merging the first wheel detection data and second wheel detection data, those first wheel detection data and second wheel detection data which are each contained in a common wheel gate are then merged.

[0035] The wheel gates are preferably generated based on a geometric model, derived from a given geometry of the target vehicle and scaled based on the target vehicle dimensions. Alternatively or additionally, the wheel gates can be generated dynamically based on the first wheel detection data and / or the second wheel detection data, with the dimensioning of the wheel gates being determined based on a given wheel diameter.

[0036] In a further advantageous embodiment, the parameter of the target vehicle is fed to a tracking filter for tracking the target vehicle.

[0037] In a further advantageous embodiment, the parameters of the target vehicle are fed to an unscented Kalman filter as a tracking filter, by means of which the current position of the target vehicle is estimated based on the parameter. This is a nonlinear Kalman filter variant based on model-based prediction with subsequent correction based on available measurement data.

[0038] In a further advantageous embodiment, the parameter of the target vehicle is fed to several different unscented Kalman filters as tracking filters, each of which is based on a different motion model. It is provided that the several different unscented Kalman filters are combined with one another and, in particular, each is weighted. For this purpose, the IMM (Interacting Multiple Model) method, known to those skilled in the art, can preferably be used. The IMM method defines an interaction between several unscented Kalman filters, which process the same measurement data in parallel.

[0039] By providing and combining such several different unscented Kalman filters as tracking filters, each based on a different movement model, different possible situations can be advantageously addressed and taken into account accordingly, so that a flexible adaptation to a specific situation can be achieved in an advantageous manner.

[0040] Unscented Kalman filters are advantageously used, which are based on one of the following four motion models: Constant Velocity Motion Model Constant Acceleration Motion Model Constant Steering Angle and Velocity Model Motion model with explicit sideslip angle estimation and non-linear single-track model.

[0041] In particular, at least four unscented Kalman filters are used, wherein each of the four unscented Kalman filters is based on a specific motion model of the aforementioned motion models, so that all four aforementioned motion models are represented.

[0042] In a further advantageous embodiment, the parameter of the target vehicle is a geometric center of gravity of the target vehicle. Using such a geometric center of gravity, the target vehicle's position and / or course can be accurately characterized and tracked, particularly in the immediate vicinity of the motor vehicle. For example, compared to a radar reflection point on the surface or body of the target vehicle, which "wanders" during a driving maneuver, the geometric center of gravity is deterministic. Even if the target vehicle is no longer completely within the detection range of the environmental sensors, in particular, is at least partially located in a so-called blind spot, precise tracking of the target vehicle based on its geometric center of gravity is still possible.

[0043] The geometric center of gravity is determined from the fused first and second wheel detection data and thus advantageously from the fused position data of at least one wheel or wheel center of gravity, in particular from the fused position data of all identified wheels. However, the geometric center of gravity can also be determined from the fused position data and the speed data of the target vehicle at the position of the wheel center of gravity of at least one wheel, in particular from the fused position data and the speed data of the target vehicle at the corresponding positions of the wheel centers of gravity of all identified wheels. The geometric center of gravity can be calculated or estimated from the corresponding data, for example, using a model-based approach.

[0044] Advantageously, in addition to the center of gravity of the target vehicle, a yaw rate and / or a speed of the target vehicle, preferably a relative speed between the motor vehicle and the target vehicle, are determined from the corresponding data of one or more wheels as further parameters of the target vehicle. The yaw rate is advantageously determined based on the speed of the target vehicle at the position of the center of gravity of at least one wheel, in particular based on the speeds of the target vehicle at the corresponding positions of the centers of gravity of all identified wheels. Knowledge of these parameters further increases the accuracy of tracking the target vehicle and, based on this, the accuracy of determining and / or determining collision-relevant data.

[0045] The second environmental sensor can be embodied, for example, as a camera or as a (further) radar sensor. The explanations explained for the first environmental sensor embodied as a radar sensor apply accordingly to the use of the (further) radar sensor as a second environmental sensor.

[0046] In a further advantageous embodiment, however, the second environmental sensor is designed as a lidar sensor, wherein the sensor data comprises lidar reflection points. Using a lidar sensor, distant objects, such as the distant target vehicle to be tracked, are typically illuminated by a laser or other light source, and the resulting lidar reflection points are recorded.

[0047] In this case, lidar reflection points of the target vehicle can first be selected from a recorded set of lidar reflection points, i.e. those lidar reflection points can be identified which are assigned to the remote target vehicle or originate from it.

[0048] In an advantageous embodiment, identifying a wheel of the target vehicle from the sensor data of the lidar sensor comprises selecting lidar reflection points of the target vehicle with a predetermined maximum height above ground, wherein a wheel is identified based only on the selected lidar reflection points. A wheel is the only component of a vehicle that touches the ground, so lidar reflection points of the target vehicle that have a predetermined maximum height above ground can only originate from a wheel. This enables accurate and reliable identification of a wheel. Furthermore, further processing can be considerably simplified, since only a very limited number of lidar reflection points need to be processed.

[0049] In a further advantageous embodiment, the provision of second wheel detection data comprises the following steps: Identifying selected lidar reflection points that have at least a predetermined number of neighboring lidar reflection points within a predetermined distance as core points, identifying selected lidar reflection points that are at most the predetermined distance from at least one core point, but have less than the predetermined number of neighboring lidar reflection points within the predetermined distance, as edge points, recognizing a respective segment as an area of ​​core points surrounded by edge points, wherein each segment is assigned to a tire tread or a tire sidewall, fitting a respective line section through each segment, determining a respective perpendicular to each of the lines centrally in the respective line section, determining a wheel center point as the intersection point of these perpendiculars, and providing the position data of the wheel center point as second wheel detection data.

[0050] Such an approach of classifying core points and edge points and detecting segments based on them has proven to be a reliable and fast way of detecting segments for typical use cases. It should be understood that lidar reflection points that do not meet the criteria for core points or edge points can be referred to as noise points. A tire tread and a tire sidewall are typical features of a wheel that cause a significant number of lidar reflection points and, accordingly, lead to numerous detected lidar reflection points. Determining the wheel center is based on the assumption that a wheel center is typically located at an intersection of the aforementioned perpendiculars. In particular, it can also be located above such an intersection.The position data are represented in particular by an x-value and a y-value in a reference coordinate system of the motor vehicle.

[0051] Advantageously, the corresponding wheel center point is determined based on a wheel model and / or other information for a wheel of the target vehicle to which both a section of tire tread and a section of tire sidewall have not been fitted. For example, it can be assumed that the typically four wheels of a target vehicle are of the same design. Thus, for example, distances detected for one wheel can be used to determine distances or the wheel center point for another wheel for which not enough lidar reflection points were recorded to separately determine the wheel center point. This can occur, for example, if the corresponding wheel is obscured by another wheel or other objects.

[0052] In a further advantageous embodiment, the provision of second wheel detection data further comprises the following steps: Determining a wheel steering angle of the target vehicle based on the fitted routes, and providing the wheel steering angle as second wheel detection data.

[0053] Both the tire tread and the tire sidewall typically depend directly on the wheel steering angle. It has been recognized that this relationship can be advantageously used to determine the wheel steering angle, i.e., for example, an angle of steered wheels relative to other elements of a body or a longitudinal axis of the remote target vehicle. Knowledge of the wheel steering angle of the target vehicle further increases the accuracy of tracking the target vehicle and, based on this, the accuracy of determining collision-relevant data, since this allows the current course of the target vehicle desired by the driver of the target vehicle or an autonomous vehicle control system to be derived.

[0054] In a further advantageous embodiment, the method further comprises the step of determining collision-relevant data based on the tracking of the target vehicle using the parameter. The determination can in particular also include defining a wheel steering angle, advantageously provided as second wheel detection data, and / or a determined parameter, such as advantageously yaw rate and / or speed, of the target vehicle as collision-relevant data. Such collision-relevant data can be used, particularly in a pre-crash phase, for the targeted activation of passive safety systems, such as, for example, tensioning a seatbelt or deploying an airbag.It is also possible to intervene in the control system of the motor vehicle carrying out the procedure, for example to carry out an evasive maneuver or to ensure that the collision occurs at a less critical point on the motor vehicle.

[0055] The present invention further comprises a collision detection device for a motor vehicle for detecting an impending collision with a remote target vehicle, comprising a first environmental sensor designed as a radar sensor and a second environmental sensor, as well as a control unit. The control unit is configured to track the remote target vehicle in an area surrounding the motor vehicle using the method according to the invention.

[0056] The advantages and preferred embodiments described for the method according to the invention also apply accordingly to the collision detection device according to the invention.

[0057] Embodiments of the invention are explained in more detail below with reference to a drawing. In the drawings: Fig. 1 shows a schematic representation of a motor vehicle for carrying out the method according to the invention together with a remote target vehicle, Fig. 2 shows a flow diagram of a method for tracking the remote target vehicle in an area surrounding the motor vehicle according to Fig. 1 , Fig. 3 in a schematic representation the provision of second wheel detection data of the target vehicle according to Fig. 1 , and Fig. 4 a wheel gating of the target vehicle according to Fig. 1 .

[0058] Corresponding parts are always provided with the same reference symbols in all figures.

[0059] Fig. 1 shows schematically a motor vehicle 1 and a remote target vehicle 2. The target vehicle 2 comprises a body 3 and four wheels 4a, 4b, 4c, 4d. In the Fig. 1In the illustration shown, the motor vehicle 1 and the distant target vehicle 2 are moving towards each other in such a way that a substantially frontal collision is imminent.

[0060] The motor vehicle 1 has a collision detection device for detecting an impending collision with the remote target vehicle 2. The collision detection device comprises a first environmental sensor 5 embodied as a radar sensor and a second environmental sensor 6 embodied as a lidar sensor, as well as a control unit that executes a method 100 for tracking the remote target vehicle 2. The radar sensor 5 is used in a fast mode, in which the data acquisition time is reduced by reducing the number of chirps per frame in order to achieve optimal cycle times. The radar sensor 5 and the lidar sensor 6 typically communicate with the control unit and make their respective data available to the control unit via a motor vehicle bus; this communication could also be wireless.

[0061] Radar sensor 5 and lidar sensor 6 are oriented in a forward direction and aligned for forward detection of the surroundings and target vehicle. The detection range 7 of radar sensor 5 and the detection range 8 of lidar sensor 6 overlap such that at least a partial area of ​​target vehicle 2 is detected by both surrounding sensors 5, 6. However, the method can also be applied accordingly for a sideways and / or rearward orientation.

[0062] Fig. 2 shows a flowchart of a method 100 for tracking a remote target vehicle 2 in an area surrounding a motor vehicle 1 according to Fig. 1 . The procedure is described below with reference to the situation in which a collision is imminent.

[0063] In a step 101, radar sensor data from radar sensor 5 and sensor data from lidar sensor 6 comprising lidar reflection points 9 are first provided. Radar reflection points are derived from the radar sensor data, which determine a position defined by a distance and an azimuth angle, and a Doppler velocity of points on the target vehicle 2. For this purpose, a three-dimensional Fourier spectrum of radar reflection points with distance, azimuth angle, and Doppler velocity is first generated based on the radar sensor data using a Fourier transformation. The radar reflection points are then separated from the noise in the radar sensor data using a CFAR filter in order to identify the radar reflection points reflected by the target vehicle 2 and to distinguish them from the signal noise contained in the radar sensor data.A clustering method designed as a DBSCAN method is then used to identify those radar reflection points that are reflected by the target vehicle 2.

[0064] In addition, lidar reflection points 9 of the target vehicle 2 are selected from a recorded set of lidar reflection points 9, i.e. those lidar reflection points 9 are identified which are assigned to the remote target vehicle 2 or originate from it.

[0065] In a step 102, one or more wheels 4a, 4b, 4c, 4d of the target vehicle 2 are then identified from the radar reflection points. This utilizes the particular movement characteristic of a (rotating) wheel 4a, 4b, 4c, 4d of the target vehicle 2, as well as the fact that the movement of a wheel 4a, 4b, 4c, 4d deviates from the movement of the body 3 of the target vehicle 2 or rotates in addition to the movement of the body 3 of the target vehicle 2. This generates an evaluable Doppler modulation around the main Doppler shift based on the micro-Doppler effect. For example, a wheel center point moves at the same speed as the body 3 of the target vehicle 2. A point located centrally on the outer radius of the wheel 4a, 4b, 4c, 4d moves at approximately twice the speed of the body 3 of the target vehicle 2 due to the rotation of the wheel 4a, 4b, 4c, 4d.In contrast, the speed of a point located at the bottom center of the outer radius of the wheel 4a, 4b, 4c, 4d is zero or slightly greater than zero due to the rotation of the wheel 4a, 4b, 4c, 4d.

[0066] The different speeds of the points of a wheel 4a, 4b, 4c, 4d thus result in different Doppler speeds of the corresponding radar reflection points, some of which differ considerably from one another and which (particularly apart from the Doppler speed of the radar reflection points corresponding to the wheel center point) also differ from the Doppler speed of the radar reflection points of the body 3. This variance in the Doppler speeds of the radar reflection points is used to identify one or more wheels 4a, 4b, 4c, 4d of the target vehicle 2.

[0067] For this purpose, a uniformly sized two-dimensional window with an extension in the distance dimension and the azimuth angle dimension is placed around each radar reflection point, and a sum of the variances of the Doppler velocities of all radar reflection points contained in the window is determined. For this purpose, an arithmetic mean is first determined from the Doppler velocities of all radar reflection points contained in the window. Then, the variance of the Doppler velocity of a radar reflection point is calculated as the difference between the Doppler velocity of the radar reflection point and the mean value for all radar reflection points contained in the window. Subsequently, the sum of the determined variances of the Doppler velocities of all radar reflection points contained in the window is calculated and assigned to the corresponding radar reflection point.A radar reflection point whose assigned sum is greater than a predetermined threshold is determined as a point of a wheel 4a, 4b, 4c, 4d.

[0068] Subsequently, the radar reflection points determined as a point of a wheel 4a, 4b, 4c, 4d of the target vehicle 2 are assigned to a wheel cluster by applying a clustering method designed as a DBSCAN method, and the wheel cluster is determined as a wheel 4a, 4b, 4c, 4d of the target vehicle 2.

[0069] In this way, one or more wheels 4a, 4b, 4c, 4d of the target vehicle can be identified extremely reliably.

[0070] In a step 103, first wheel detection data is then provided based on one or more wheels 4a, 4b, 4c, 4d identified from the radar reflection points. For this purpose, for each identified wheel 4a, 4b, 4c, 4d, the position data of a respective wheel center of gravity as well as the respective speed data of the target vehicle at the positions of the corresponding wheel centers are determined based on the radar reflection points assigned to a corresponding wheel cluster and provided as first wheel detection data for the respective wheel 4a, 4b, 4c, 4d.

[0071] The wheel center of gravity or the position data of the wheel center of gravity represent the position of the corresponding wheel 4a, 4b, 4c, 4d. The position data are each represented by an x-value and a y-value in a reference coordinate system of the motor vehicle 1 and are determined by averaging the corresponding x-values ​​and y-values ​​of the radar reflection points assigned to the corresponding wheel cluster.

[0072] To determine the speed data of the target vehicle at the positions of the corresponding wheel centers of gravity, for each identified wheel 4a, 4b, 4c, 4d, an azimuth angle value of the radar reflection point closest to the wheel center of gravity is first determined. Then, the radar reflection points that have the determined azimuth angle value as an azimuth angle value are grouped together. The radar reflection point of the group that has a relatively highest intensity value within the group and that has not been identified as a point of the corresponding wheel 4a, 4b, 4c, 4d is determined. The Doppler velocity of the radar reflection point thus determined is provided as the speed of the target vehicle 2 at the position of the corresponding wheel center of gravity as (further) first wheel detection data.

[0073] In a step 104, which can run parallel to step 102, one or more wheels 4a, 4b, 4c, 4d of the target vehicle 2 are identified from the lidar reflection points 9 of the lidar sensor 6. For this purpose, only the first lidar reflection points 9 of the target vehicle 2 that are above the ground or are only a maximum of a predetermined distance above the ground are assigned to the wheels 4a, 4b, 4c, 4d. The wheels 4a, 4b, 4c, 4d are the only components of a target vehicle 2 that touch the ground, so that lidar reflection points 9 of the target vehicle 2 that have a predetermined maximum height above the ground can only originate from one wheel 4a, 4b, 4c, 4d. This enables precise and reliable identification of one or more wheels 4a, 4b, 4c, 4d. In addition, further processing can be considerably simplified since only a relatively limited number of lidar reflection points 9 need to be further processed.

[0074] In a step 105, second wheel detection data are then provided based on one or more wheels 4a, 4b, 4c, 4d identified from the lidar reflection points 9. This is shown in Fig. 3 based on a schematic representation of the target vehicle 2 according to Fig. 1 illustrated as an example for a wheel 4b.

[0075] On the wheel 4b, the corresponding lidar reflection points 9 are clustered or segmented using a density-based method. Fig. 3Only individual lidar reflection points are provided with the reference number 9 in order not to overload the image. For example, lidar reflection points 9 that are closely located and separated from one another by a certain point-poor space can be considered. Lidar reflection points 9 that have at least a predetermined number of other lidar reflection points 9 within a previously determined distance or a predetermined spacing are core points. Core points whose spacing is smaller than the predetermined spacing typically belong to the same cluster or the same segment 10. Lidar reflection points 9 that do not meet this criterion but are close to a cluster or segment 10 are classified as edge points. Lidar reflection points 9 that do not meet both criteria are considered noise points.Thus, two segments 10 are present on the wheel 4b, one segment 10 having two lidar reflection points 9 and another segment 10 having four lidar reflection points 9.

[0076] A line or segment 11 is now fitted into the segmented lidar reflection points 9, which represents the position and angle of the tire tread of the wheel 4b, and another segment 11 is fitted, which represents the tire sidewall. The tire tread is in Fig. 3 indicated by two lidar reflection points 9, whereas the tire sidewall is indicated by four lidar reflection points 9.

[0077] These two lines or segments 11 are perpendicular to each other. By setting up a perpendicular along the lines 11, an intersection point of the perpendiculars is established, which represents a wheel center 12 of the wheel 4b. The position data of the wheel center 12, represented in particular by an x-value and a y-value in the reference coordinate system of the motor vehicle 1, are provided as second position data. Furthermore, a wheel steering angle can be determined based on the fitted segments 11 and provided as second wheel detection data.

[0078] In a step 106 of method 100, the first wheel detection data and the second wheel detection data are then combined. For this purpose, the first wheel detection data and the second wheel detection data are each initially assigned to a target vehicle axle and a target vehicle side.

[0079] For this assignment, wheel gates 13a, 13b, 13c, 13d are generated, to which the first wheel detection data and the second wheel detection data are assigned based on the corresponding position data of the first wheel detection data and the second wheel detection data.

[0080] Such wheel gating of the target vehicle 2 is in Fig. 4For the target vehicle 2, which is a two-track, two-axle vehicle, four wheel gates 13a, 13b, 13c, 13d are generated geometrically based on a model, derived from a predefined geometry of the target vehicle 2 and scaled based on the target vehicle dimensions. Alternatively or additionally, the wheel gates 13a, 13b, 13c, 13d can be generated dynamically based on the first wheel detection data and / or the second wheel detection data, with the dimensioning of the wheel gates 13a, 13b, 13c, 13d being determined based on a predefined wheel diameter. Thus, there is a wheel gate front-right 13a, front-left 13b, rear-right 13c and rear-left 13d, wherein the first wheel detection data and the second wheel detection data are each assigned to the corresponding wheel gate 13a, 13b, 13c, 13d.

[0081] Subsequently, based on a covariance intersection algorithm, the first wheel detection data and second wheel detection data each contained in a common wheel gate 13a, 13b, 13c, 13d are merged. The use of a covariance intersection algorithm in the step of merging the first and second wheel detection data assigned to a common wheel gate 13a, 13b, 13c, 13d ensures a convergent fusion of the first wheel detection data with the second wheel detection data.

[0082] In a step 107 of the method 100, the geometric center of gravity 14 of the target vehicle 2 is then determined as a parameter of the target vehicle 2 based on the combined first and second wheel detection data. The geometric center of gravity 14 is calculated or estimated, in particular, from the fused position data of all identified wheels 4a, 4b, 4c, 4d, for example, based on a model and / or using a so-called motion state transformation. Using the geometric center of gravity 14, the target vehicle 2 can be well characterized and tracked in its position and / or course, particularly in the immediate vicinity of the motor vehicle 1. For example, compared to a specific radar reflection point of the body 3 of the target vehicle 2, which "wanders" during a driving maneuver, the geometric center of gravity 14 is deterministic.Even in the event that the target vehicle 2 is no longer completely within the detection range of the environmental sensors 5, 6, in particular is at least partially located in a so-called blind spot, precise tracking of the target vehicle 2 based on its geometric center of gravity 14 is still possible.

[0083] In addition, a yaw rate and a speed of the target vehicle 2 are determined as further parameters of the target vehicle 2 from the fused position data of all identified wheels 4a, 4b, 4c, 4d and speed data at the corresponding positions of the wheel centers of gravity of all identified wheels 4a, 4b, 4c, 4d.

[0084] In a step 108, the geometric center of gravity 14 of the target vehicle 2 is fed to a tracking filter for tracking the target vehicle 2, by means of which the current position of the target vehicle 2 is estimated based on the geometric center of gravity 14. The tracking filter comprises four different unscented Kalman filters, each based on a different motion model. The four different unscented Kalman filters are combined with each other and each weighted. For this purpose, the IMM method (Interacting Multiple Model method) is used, which defines an interaction between the four unscented Kalman filters, which process the geometric center of gravity 14 of the target vehicle 2 in parallel. The four unscented Kalman filters are based on the following four motion models: Constant Velocity Motion Model Constant Acceleration Motion Model Constant Steering Angle and Velocity Model Motion model with explicit sideslip angle estimation and non-linear single-track model.

[0085] By providing and combining these four different unscented Kalman filters as tracking filters, different possible situations can be addressed and taken into account accordingly, so that a flexible adaptation to a specific situation can be achieved and thus the most accurate tracking possible.

[0086] In a step 109, collision-relevant data, such as in particular a yaw rate, a speed, and a wheel steering angle of the target vehicle 2, are determined based on the tracking of the target vehicle 2 using the geometric center of gravity 14. Such determination is also to be understood as defining as collision-relevant data the wheel steering angle provided as second wheel detection data and the yaw rate and speed of the target vehicle 2 determined as parameters. Such collision-relevant data can be used, particularly in a pre-crash phase, for the targeted activation of passive safety systems of the motor vehicle 1, such as tensioning a seatbelt or deploying an airbag.It is also possible to intervene in the control of the motor vehicle 1 executing the method 100, for example to carry out an evasive maneuver or to ensure that the collision occurs at a less critical point on the motor vehicle 1. List of reference symbols

[0087] 1 Motor vehicle 2 Target vehicle 3 Body 4a, 4b, 4c, 4d Wheel 5 First environmental sensor (radar sensor) 6 Second environmental sensor (lidar sensor) 7 Radar sensor detection range 8 Lidar sensor detection range 9 Lidar reflection point 10 Segment 11 Route 12 Wheel center point 13a Front-right wheel gate 13b Front-left wheel gate 13c Rear-right wheel gate 13d Rear-left wheel gate 14 Geometric center of gravity 100 Method for tracking a remote target vehicle 101 Providing radar sensor data and target vehicle sensor data 102 Identifying a wheel of the target vehicle from the radar reflection points 103 Providing first wheel detection data 104 Identifying a wheel of the target vehicle from the lidar reflection points 105 Providing second wheel detection data 106Combining the first and second wheel detection data 107Determining a parameter of the target vehicle 108Feeding a parameter to a tracking filter 109Determining collision-relevant data

Claims

1. Method (100) for tracking a distant target vehicle (2) in an area surrounding a motor vehicle (1) by means of a collision detection device of the motor vehicle (1) for detecting an impending collision with the target vehicle (2) by fusing radar sensor data from a first surroundings sensor (5) designed as a radar sensor with sensor data from a second surroundings sensor (6) of the collision detection device, wherein the radar sensor (5) and the second surroundings sensor (6) each have capture ranges (7, 8) which overlap in such a way that at least one section of the target vehicle (2) is captured by both surroundings sensors (5, 6), and wherein the method (100) comprises the following steps: - providing (101) radar sensor data and sensor data relating to the target vehicle (2) by means of the radar sensor (5) and the second surroundings sensor (6), wherein radar reflection points are derived from the radar sensor data and determine a position determined by a distance and an azimuth angle and a Doppler speed of points on the target vehicle (2), - identifying (101) a wheel (4a, 4b, 4c, 4d) of the target vehicle (2) from the radar reflection points, - providing (103) first wheel capture data based on the wheel (4a, 4b, 4c, 4d) identified from the radar reflection points, - identifying (104) a wheel (4a, 4b, 4c, 4d) of the target vehicle (2) from the sensor data from the second surroundings sensor (6) and providing (105) second wheel capture data (12) based on the wheel (4a, 4b, 4c, 4d) identified from the sensor data, - merging (106) the first wheel capture data and the second wheel capture data, and - determining a parameter (14) of the target vehicle (2) based on the merged first and second wheel capture data, characterized in that the identification (101) of a wheel (4a, 4b, 4c, 4d) comprises placing in each case a two-dimensional window of uniform size and with an extent in the distance dimension and the azimuth angle dimension around each radar reflection point and determining a sum of the variances of the Doppler speeds of all radar reflection points contained in the window and assigning it to the corresponding radar reflection point, and determining a radar reflection point as a point of a wheel (4a, 4b, 4c, 4d) whose assigned sum is greater than a predefined threshold value.

2. Method (100) according to Claim 1, wherein the derivation of the radar reflection points from the radar sensor data comprises the following steps: - determining radar reflection points from the radar sensor data by means of Fourier transformation, - separating the radar reflection points from the noise of the radar sensor data using a CFAR filter, and - identifying the radar reflection points reflected by the target vehicle by means of a cluster method, preferably a DBSCAN method.

3. Method (100) according to Claim 1 or 2, wherein the determination of the sum of the variances of the Doppler speeds of all radar reflection points contained in a window comprises the following steps: - determining an arithmetic mean value from the Doppler speeds of all radar reflection points contained in the window, - determining a variance of the Doppler speed of a radar reflection point as the difference between the Doppler speed of the radar reflection point and the mean value in each case for all radar reflection points contained in the window, and - forming the sum of the determined variances of the Doppler speeds of all radar reflection points contained in the window.

4. Method (100) according to one of the preceding claims, wherein the identification (101) of a wheel (4a, 4b, 4c, 4d) of the target vehicle (2) from the radar reflection points further comprises the following steps: - assigning the radar reflection points determined as a point of a wheel (4a, 4b, 4c, 4d) to a wheel cluster by means of a cluster method, preferably a DBSCAN method, and - determining the wheel cluster as a wheel (4a, 4b, 4c, 4d) of the target vehicle (2).

5. Method (100) according to Claim 4, wherein the provision (103) of first wheel capture data comprises the following steps: - calculating the position data for a wheel centre of gravity based on the radar reflection points assigned to a wheel cluster, and - providing the position data for the wheel centre of gravity as first wheel capture data.

6. Method (100) according to Claim 5, wherein the provision (103) of first wheel capture data further comprises the following steps: - determining an azimuth angle value of the radar reflection point positionally closest to the wheel centre of gravity, - combining the radar reflection points having the determined azimuth angle value as the azimuth angle value in a group, - determining that radar reflection point in the group which has a relatively highest intensity value within the group and which has not been simultaneously determined as a point of a wheel (4a, 4b, 4c, 4d), - providing the Doppler speed of the determined radar reflection point as the speed of the target vehicle (2) at the position of the wheel centre of gravity as first wheel capture data.

7. Method (100) according to one of the preceding claims, wherein the merging (106) of the first wheel capture data and the second wheel capture data comprises the following steps: - respectively assigning the first wheel capture data and the second wheel capture data to a target vehicle axis and to a target vehicle side, - merging the first wheel capture data and second wheel capture data assigned to a common target vehicle axis and target vehicle side based on a covariance intersection algorithm.

8. Method (100) according to one of the preceding claims, further comprising the step of supplying (108) the parameter (14) to a tracking filter for tracking the target vehicle (2).

9. Method (100) according to Claim 8, wherein the parameter (14) of the target vehicle (2) is supplied to an unscented Kalman filter as a tracking filter, by means of which the current position of the target vehicle (2) is estimated based on the parameter (14).

10. Method (100) according to Claim 9, wherein the parameter (14) of the target vehicle (2) is supplied to multiple different unscented Kalman filters as tracking filters, wherein the multiple different unscented Kalman filters are each based on a different motion model.

11. Method (100) according to one of the preceding claims, wherein the parameter (14) of the target vehicle (2) is a geometric centre of gravity of the target vehicle (2).

12. Method (100) according to one of the preceding claims, wherein the second surroundings sensor (6) is designed as a lidar sensor, and wherein the sensor data comprise lidar reflection points (9).

13. Method (100) according to Claim 12, wherein the identification of a wheel (4a, 4b, 4c, 4d) of the target vehicle (2) from the sensor data from the lidar sensor (6) comprises selecting lidar reflection points (9) of the target vehicle (2) with a predefined maximum height above ground, and wherein a wheel (4a, 4b, 4c, 4d) is identified only based on the selected lidar reflection points (9).

14. Method (100) according to Claim 13, wherein the provision (105) of second wheel capture data comprises the following steps: - identifying selected lidar reflection points (9), which have at least a predefined number of neighbouring lidar reflection points within a predefined distance, as core points, - identifying selected lidar reflection points (9), which have at most the predefined distance from at least one core point, but have less than the predefined number of neighbouring lidar reflection points within the predefined distance, as edge points, - identifying a respective segment (10) as an area of core points surrounded by edge points, each segment (10) being assigned to a tyre tread or a tyre sidewall, - fitting a respective route (11) through each segment (10), - determining a respective perpendicular to each of the routes in the centre of the respective route (11), - determining a wheel centre point (12) as the point of intersection of these perpendiculars, and - providing the position data relating to the wheel centre point (12) as second wheel capture data.

15. Method (100) according to Claim 14, wherein the provision (105) of second wheel capture data further comprises the following steps: - determining a wheel steering angle of the target vehicle (2) based on the fitted routes (11), and - providing the wheel steering angle as second wheel capture data.

16. Method (100) according to one of Claims 8 to 15, further comprising the step of determining (109) collision-relevant data based on the tracking of the target vehicle (2) based on the parameter (14).

17. Collision detection device for a motor vehicle (1) for detecting an impending collision with a distant target vehicle (2), having a first surroundings sensor (5) designed as a radar sensor and a second surroundings sensor (6) and a control unit, wherein the control unit is prepared to track the distant target vehicle (2) in an area surrounding the motor vehicle (1) by means of a method (100) according to at least one of the preceding claims.