Method and system for detecting false detections

Abstract

The invention relates to a model-based method for determining false detections in radar detections in radar data, in connection with cross paths that arise due to different transmission and reception angles of paths for a radar target, in which a radar cube with range Doppler cells is generated using Fourier transformation on the basis of the radar data, wherein first a range and then a Doppler Fourier transformation is carried out so that channel data is available for each range Doppler cell, wherein the determination of false detections is carried out via the channel data of several adjacent range Doppler cells by using multidimensional models for multipath propagation that span these range Doppler cells to determine the detection parameters range, Doppler and angle of the detections, wherein the following models are used as multipath propagation models: - Multi-path model comprising a direct path, an indirect path and an associated cross-path for a radar target; and / or - Cross-path model, which includes only a cross-path for a radar target (and no direct and indirect path); and / or - Multiple single target models, which include multiple direct independent paths or multiple independent radar targets.

Description

[0001] The present invention relates to a method and a system for detecting false detections. Technological background

[0002] Wireless transmissions via radio waves, such as radar transmissions, are susceptible to propagation effects such as sensor blockage, scattering, or multipath. In monostatic radar applications, multipath propagation leads to so-called "false positives," i.e., incorrectly identified targets or detections. In radar application scenarios in the automotive sector, multipath propagation often occurs on smooth surfaces, such as guardrails. The general propagation paths are described in Fig. 1, where a total of four different propagation paths are shown and in the table in Fig. 1 shall be designated accordingly.

[0003] Depending on the radar sensor's detection algorithm and its resolution, the four different propagation paths within such a multipath scenario can lead to target detection. Propagation via the indirect path and the two cross paths, in particular, can lead to misclassified targets (i.e., false detections or false positives).

[0004] In multipath propagation, cross paths can also occur. A cross path has a different propagation path on the outward and return paths. It thus represents a mixture of direct and indirect propagation paths. Typically, there is a first cross path that uses the direct propagation path on the outward path but the indirect propagation path on the return path, and a second cross path that, inversely, uses the indirect propagation path on the outward path but the direct propagation path on the return path. A cross path is thus characterized by the fact that the angle of emergence and the angle of incidence of the radar signal are different. State of the art

[0005] DE 10 2012 105 582 A1 discloses a method for a radar system that evaluates multiple beams to determine the angular position of objects in at least one coordinate. A high-resolution method for separating multiple reflection points based on a cost function is repeatedly executed. The calculation of the cost function is then split into a part that is independent of the input data and a part that is dependent on the input data. The part that is independent of the input data is determined and stored in advance at predetermined support points for parameters of the reflection points.

[0006] Furthermore, DE 10 2016 215 509 A1 describes a method for operating a radar system in a vehicle, in which a mirror surface is first detected and the minimum distance from an object to the mirror surface is determined, wherein the object is classified as a mirror object if the object and the vehicle are located on different sides of the mirror surface

[0007] Furthermore, in "Automotive MIMO Radar Angle Estimation in the Presence of Multipath" (October 2017; DOI: 10.23919 / EURAD.2017.8249152), Engels et al. describe MIMO schemes used in automotive radar to improve resolution and signal-to-noise ratio. Practical scenarios that typically require high-resolution angle estimation include horizontal or vertical multipath. In these situations, the standard two-target model can no longer be applied when using MIMO schemes, and model-based angle estimation must be extended. To this end, they disclose two-target and single-target signal models in the presence of multipath, defining a multipath scenario and multipath within a range / Doppler cell, respectively. Corresponding maximum likelihood (ML) estimators are also described.

[0008] From DE 10 2021 211 989 B3 a method is known for associating a detection of a real target detected by a radar sensor with its mirror target on the basis of a detected cross path, in which the angle of incidence and reflection of the cross path, which arises due to a reflection on a mirror surface, as well as its length are determined, wherein based on the angle of incidence and reflection and the length of the cross path, radar detections are grouped by respectively assigning a radar detection to a group of a set of several groups.Furthermore, at least one radar detection of a first group is associated with a radar detection of a second group, wherein the radar detection of the first group is arranged in an angular range around the angle of incidence and the radar detection of the second group is arranged in an angular range around the angle of reflection, and the first group of radar detections has a radial distance from the radar sensor less than the length of the cross path and the second group of radar detections has a radial distance from the radar sensor greater than the length of the cross path, or vice versa, and wherein the sum of the radial distances of the associated radar detections from the radar sensor is substantially equal to twice the length of the cross path. As a result, information can then be provided that the associated radar detections are a real target and an associated mirror target.

[0009] The values ​​used in DE 10 2021 211 989 B3 are based on detection estimates. Especially in MIMO (Multiple Input Multiple Output) setups, the azimuth estimates are distorted. Furthermore, not all components of the multipath scenario are detectable, which prevents part of the algorithm from working. In automotive scenarios where the target is at a greater distance, for example, the mirror target may no longer be detectable because it is too close to one of the crossing paths and has lower power. Object of the present invention

[0010] Based on this, it is the object of the invention to provide a method for detecting false detections in which the detection accuracy can be improved and the number of false detections can be reduced. Inventive solution

[0011] The above object is achieved by the combination of features of claim 1 and the subordinate claim. Advantageous embodiments of the invention are claimed in the subclaims.

[0012] The present invention relates to a (model-based) method for determining false detections in radar data, in conjunction with cross paths that arise due to different transmission and reception angles of paths for a radar target. In this method, a radar cube with range Doppler cells is generated from the radar data using a Fourier transformation. First, a range and then a Doppler Fourier transformation are performed, so that channel data are available for each range Doppler cell. False detections are determined using the channel data of several neighboring range Doppler cells (neighboring cells). Multidimensional multipath propagation models spanning these range Doppler cells are used to determine the detection parameters range, Doppler, and angle of the detections. Furthermore, the following models are used as multipath propagation models: - Multi-path model comprising a direct path, an indirect path and an associated cross-path for a radar target; and / or - Cross-path model, which includes only a cross-path for a radar target (and no direct and indirect path); and / or - Multiple single target models, which include multiple direct independent paths or multiple independent radar targets.

[0013] The present invention enables the reliable detection of a multipath scenario based on raw radar data, whereby increased accuracy of the detected values ​​can be achieved based on physically related models and the number of falsely detected targets or false detections is reduced.

[0014] Preferably, the number of detection parameters for the multipath model is reduced so that both the rank and Doppler values ​​of the cross paths correspond to the mean of the corresponding values ​​of the indirect and direct paths. For example, the geometric relationship br,m=br,c+r=br,d+2r bd,m=bd,c+d=bd,d+2d The geometric relationship is based on the cross paths lying between the corresponding values ​​of the direct and mirror target in range and Doppler.

[0015] Pre-filtering can be advantageously carried out by using a beamformer to determine at least two detections with a maximum power difference, whereby the range / Doppler values ​​of the two detections must not exceed a threshold value in the difference of the respective value, or the difference between range and Doppler of the detections found should be below a certain threshold value.

[0016] Furthermore, the determination of the detection parameters range, Doppler, and angle can be divided into two parts: first, the angle values ​​of the models are estimated, followed by the range-Doppler values. The range-Doppler estimation can also be divided into two separate parts.

[0017] It is also useful to display both cross paths with a common amplitude.

[0018] Furthermore, the present invention comprises a system for determining false detections in radar detections in radar data, in connection with cross paths that arise due to different transmission and reception angles of paths for a radar target, wherein the system comprises at least one radar sensor for detecting the surroundings of a vehicle and a control unit for processing the information provided by the radar sensor, wherein the control unit is designed to carry out a method according to one of the preceding claims. Description of the invention using an embodiment

[0019] In Fig.Figure 2 shows a simplified schematic representation (of a section) of a radar cube comprising a type of multidimensional data structure in which, for example, the raw radar data can be stored in a spatially discrete and preferably structured manner. This shows the current range-Doppler cell (or the processing cell of interest) as well as the surrounding (eight) neighboring cells, including all channels or channel data for a range-Doppler cell.

[0020] A key point of the present invention is that the crossing paths with other models are differentiated using the channel data (so-called "channel data", which are the received samples for a specific receive / transmit antenna combination after range and Doppler Fast Fourier Transform (FFT) processing). Assuming that the channel data are from a range of N r times N dprocessing cells around a specific processing cell, the sampled data for L single targets can be stored as z rd|st be described: zrd|st=[(Br⊗Bd)(Ar(br)⋄Ad(bd))]⋄Aϕ(bϕ)α+n where ⊗ is the Kronecker product and ◊ the Kathri-Rao product, ie a column-wise Kronecker product. The matrices B r and B d provide the Fourier transform and windowing around the processing cell of interest, while A r (b r ), A d (b d ), A ϕ (b ϕ ) denote the steering matrices in range, Doppler, and azimuth angles, respectively, where each column of the matrices denotes a steering vector per target in its corresponding range. The vectors b r , b d and b ϕhave the length L and denote the position of the L-targets with respect to range, Doppler, or azimuth values, and α the corresponding amplitude vector. Furthermore, in equation (1) a virtual array within A ϕ (b ϕ ) is used according to the monostatic MIMO (Multiple-Input Multiple-Output) concept. The noise vector n is circularly symmetric, complex-normal, and distributed as a zero-mean, usually with a covariance matrix: C=σ2[(Br⊗Bd⊗INϕ)(Br⊗Bd⊗INϕ)H], assuming white, circularly symmetric complex normally distributed noise samples with variance σ 2 at the input before FFT processing where I Nϕ for the identity matrix of size N ϕ × N ϕ A total of 3·L nonlinear parameters must be estimated. Assuming that only the two cross paths within the N r · N d -Processing cells can process the channel data rd|cp be described by zrd|cp=[(Br⊗Bd)(Ar(br,c)⋄(Ar(br,c)⋄Ad(bd,c))]︸[] with the distance value b r,c , the Doppler value b d,c and the azimuth angles b ϕ = [b ϕ1 , b ϕ2 ]. The control matrix A T,R,ϕ (b ϕ ) takes into account the different angle of arrival (AoA - angle of arrival) and angle of departure (AoD - angle-of-departure) by AT,R,ϕ(bϕ)=[aT(bϕ1)⊗aR(bϕ2),aT(bϕ2)⊗aR(bϕ1)], where a T (b) and a R (b) represent the steering vectors of the transmitting and receiving antenna arrays, respectively. Please note that if a T (b) and a R (b) are well characterized for the sensor, ie can be corrected, a simpler model for A T,R,ϕ (b ϕ ) can be used as: AT,R,ϕ(bϕ)=aT(bϕ1)⊗aR(bϕ2)+aT(bϕ2)⊗aR(bϕ1), where only a single amplitude needs to be used in equation (3). However, a total of four nonlinear parameters (b r,c , b d,c , b ϕ1 , b ϕ2 ) can be estimated.

[0021] Assuming that all paths of the multipath scenario are within the N r · N d -processing cells, e.g. if the target and the mirror target have a similar range and Doppler, then the channel data can be rd|mp be described by zrd|mp=[(Br⊗Bd)(Ar(br,d)⋄Ad(bd,d))]⋄aT(bϕ1)⋄aR(bϕ1), (Br⊗Bd)(Ar(br,c)⋄Ad(bd,c))⋄aT(bϕ1)⋄aR(bϕ2), (Br⊗Bd)(Ar(br,c)⋄Ad(bd,c))⋄aT(bϕ2)⋄aR(bϕ1), (Br⊗Bd)(Ar(br,m)⋄Ad(bd,m))⋄aT(bϕ2)⋄aR(bϕ2)]α+n, where the position of the target (b r,d , b d,d , b ϕ1 ) is, for the crossing lanes (b r,c , b d,c ) with the angle values ​​b ϕ1 and b ϕ2 , while the mirror at (b r,m , b d,m, b ϕ2 ) is located. Using the geometric relationships Br,m=br,c+r=br,d+2r bd,m=bd,c+d=bd,d+2d, the number of independent parameters can be reduced, ie only 6 nonlinear parameters (b r,c , b d,c , r, d, b ϕ1 , b ϕ2 ) can be estimated since all paths are interdependent.

[0022] In automotive scenarios, the "cross path" scenario represents the most critical scenario, as two additional false targets can be generated. If all four paths are significant—that is, the full multipath model, including direct, mirror, and cross paths—then only two targets will likely be generated, one at the mirror path and one at the target location. This is because the discriminatory power of the mirror path is likely higher than the cross path, and the direct path likely has higher performance than the corresponding cross path.

[0023] To distinguish between the different situations, the detection parameters can then be estimated from the corresponding models to calculate the mean squared error (MSE). The use of standard model discrimination techniques, such as the generalized likelihood ratio test (GLRT) or information criteria (IC), makes it possible to identify the correct model and mark the detections as false positive targets, false detections, or real targets using the known method from DE 10 2021 211 989 B3. Simplified determination of parameters

[0024] The following describes a simplified estimator for the described models. The estimation is divided into two parts or ranges: first, the angle values ​​of the models are estimated, followed by the range Doppler values.

[0025] In the first step or first part, the angle values ​​b φ either estimated within the area or Doppler cell with the highest power or incoherently over all N r · N d -cells. The estimation can be performed using the maximum likelihood (ML) approach, e.g., maximizing the lumped probability distribution by b^ϕ=minbϕ tr{(I−A(AHA)AH)zzH}, where A is the angle control matrix depending on b ϕ for the corresponding model and z denotes the channel data used within the estimation. With the angle estimates b̂ ϕ The complex amplitudes can be estimated using least squares (LS) for each processing cell as follows: α^i=(AHA)−1AHzi, where i = 1, ..., N r N d and z iare the channel data of the i-th cell. The covariance matrix R ϕ The estimated amplitudes can be simplified to Rϕ=σ2(AHA)−1, where σ 2 the variance of the white circularly symmetric complex normal distribution of the noise in z i Stacking the estimated complex amplitudes per cell in a matrix Γ with Γ=[α^1T⋮α^NrNdT] describes a circularly symmetric complex normally distributed random matrix whose logarithmic probability distribution is proportional to log p(Γ)∝tr{Rϕ−1(Γ−ArdW)HRrd−1(Γ−ArdW)}, where a simplified notation with A rd as a control matrix depending on the objectives within the model or b r or b d The matrix W is a diagonal matrix with complex weights for each target in the model, while the covariance matrix is ​​described over the processing cells, by Rrd=(Br⊗Bd)(Br⊗Bd)H.

[0026] The optimization in equation (12) refers to W, ie the linear parameter results Wopt=(ArdHRrd−1Ard)−1ArdHRrd−1Γ.

[0027] The use of W opt in equation (12) allows the cost function that must be minimized to be easily determined as (b^r,b^d)=minbr,bd tr{Rϕ−1ΓHRrd−1Γ −Rϕ−1ΓHRrd−1Ard(ArdHRrd−1Ard)−1ArdHRrd−1Γ}. Requirements for the evaluation

[0028] Since the described tests are comparatively complex in terms of the calculations described, they should not, for example, be calculated for each area or each Doppler cell or detection. To reduce the computational effort, the data of a processing cell should therefore be examined before the model estimation described above. The present invention can be used in particular to distinguish between "cross-only situations" or cross-path propagations, which are particularly critical with regard to false detections. In this case, at least two peaks with a maximum power difference of Δ should be found in the beamformer, the difference between the peaks found should be below a certain threshold in range and Doppler, and any peaks found should not be classified as stationary. QUOTES CONTAINED IN THE DESCRIPTION

[0000] This list of documents submitted by the applicant was generated automatically and is included solely for the convenience of the reader. This list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions. Cited patent literature

[0000] DE 10 2012 105 582 A1

[0005] DE 10 2016 215 509 A1

[0006] DE 10 2021 211 989 B3 [0008, 0009, 0023] Cited non-patent literature

[0000] ENGELS et al. in Automotive MIMO Radar Angle Estimation in the Presence of Multipath (October 2017; DOI: 10.23919 / EURAD.2017.8249152

[0007]

Claims

[1] Method for determining false detections in radar detections in radar data, in connection with cross paths that arise due to different transmission and reception angles of paths for a radar target, in which a radar cube with range Doppler cells is generated from the radar data using Fourier transformation, where first a range and then a Doppler-Fourier transformation is carried out, so that This provides channel data for each range Doppler cell, whereby false detections are determined using the channel data of several neighboring range Doppler cells by To determine the detection parameters range, Doppler and angle of the detections, multidimensional models for multipath propagation that span these range Doppler cells are used, whereby the following models are used as multipath propagation models: - Multi-path model comprising a direct path, an indirect path and an associated cross-path for a radar target; and / or - Cross-path model, which includes only a cross-path for a radar target (and no direct and indirect path); and / or - Multiple single target models, which include multiple direct independent paths or multiple independent radar targets. [2] Method according to claim 1, characterized by that the number of detection parameters for the multipath model is reduced by the geometric relationship that both the value of the rank and that of the Doppler of the cross paths correspond to the mean of the corresponding values ​​of the indirect and direct paths. [3] Method according to claim 1 or 2, characterized by, which is pre-filtered by using a beamformer to determine at least two detections with a maximum power difference of Δ, whereby the difference in rank and Doppler of the detections found should be below a certain threshold. [4] Method according to one of the preceding claims, characterized by that the determination of the detection parameters range, Doppler and angle is divided into two parts, where first the angle values ​​of the models are estimated and then the range-Doppler values ​​are estimated. [5] Method according to claim 4, characterized by that the estimation of the range Doppler values ​​is not carried out together, but is divided into two separate parts. [6] Method according to one of the preceding claims, characterized by that both cross paths are represented with a common amplitude. [7] System for determining false detections in radar detections in radar data, in connection with cross paths that arise due to different transmission and reception angles of paths for a radar target, comprising at least one radar sensor for detecting the surroundings of a vehicle and a control unit for processing the information provided by the radar sensor, wherein the control unit is designed to carry out a method according to one of the preceding claims.

Images