Method for resolving angle ambiguity in radar networks.

Abstract

The present invention relates to a method for resolving angular ambiguities in a spatially incoherent radar network (2). According to the invention, the surrounding area is scanned by a number of radar sensors (2.1 to 2.n). For each radar sensor (2.1 to 2.n) individually and independently of each other radar sensor (2.1 to 2.n), tracks (T1.1 to T1.m ... Tn.1 to Tn.x) of detected objects are generated in a state space. The tracks (T1.1 to T1.m ... Tn.1 to Tn.x) of the different radar sensors (2.1 to 2.n) are assigned to each other on condition that there is a plausibility that the tracks originate from the same object. To resolve the angular ambiguity, the tracks (T1.1 to T1.m ... Tn.1 to Tn.x) assigned to each other are fused for different variants, a respective plausibility measure is assigned to each variant and the most plausible variant is selected to resolve the angular ambiguity.

Description

[Technical field]

[0001] The present invention relates to a method for resolving angular ambiguities in spatially incoherent radar networks. [Background technology]

[0002] During automated, and especially highly automated or autonomous, operation of a vehicle, radar sensors generate a model of the surrounding space by detecting backscattering from stationary and moving objects. However, angular ambiguities can make it difficult to unambiguously determine the incoming direction of the signal and thus the position of the detected object.

[0003] US2019 / 0187268A1 discloses a method and apparatus for resolving radar angle ambiguity, in which the angular position of an object is determined from a spatial response having multiple amplitude peaks. To resolve the radar angle ambiguity, a frequency subspectrum or multi-frequency subspectrum is selected, which highlights the amplitude or phase differences in the spatial response and analyzes the irregular morphology of the response over a wide field of view to determine the angular position of the object. This gives the angular position of the object a distinct signature that the radar system can detect and use to resolve the radar angle ambiguity. The radar antenna elements are arranged in an array with a distance between them that is greater than 1 / 2 the average wavelength of the reflected radar signal used to detect the object. Summary of the Invention

[0004] The object of the present invention is to provide a new type of method for resolving angular ambiguities in spatially incoherent radar networks.

[0005] The invention achieves this object by means of a method having the features of claim 1.

[0006] Advantageous embodiments of the invention are the subject of the dependent claims.

[0007] In a method for resolving angle ambiguity in a spatially incoherent radar network according to the present invention, multiple radar sensors detect surrounding areas, and tracks of detected objects are generated in a state space for each radar sensor individually and independently of each other radar sensor. The state space is configured, for example, such that there is no need to resolve angle ambiguity therein. Furthermore, different radar sensor tracks are assigned to each other, provided that there is a plausibility that the tracks originate from the same object. The assigned tracks are fused to different variants to resolve the angle ambiguity, where a respective plausibility measure is assigned to each variant. The most plausible variant is then selected to resolve the angle ambiguity.

[0008] The disclosed method allows the angular ambiguity of object positions to be resolved by a plausibility check performed by multiple radar sensors in a radar network. This allows the number of mislocated objects, also known as ghost objects, to be reduced due to angular ambiguity, leading to a significant improvement of the surroundings model generated by the radar sensor. This allows the driving behavior to be significantly improved in the context of a system for the automatic operation of a vehicle, for example, by avoiding braking operations for objects that are not actually located at a particular position.

[0009] Unlike the current state of the art, the disclosed method is not based on frequency subspectrum and can therefore also be used with radar sensors that do not control the resolution of the angle ambiguity themselves.

[0010] Examples of the invention are explained in more detail below with reference to the drawings. [Brief description of the drawings]

[0011] [Figure 1] FIG. 1 is a schematic block diagram of an apparatus for resolving angular ambiguities in a spatially incoherent radar network. [Diagram 2]1 is a schematic diagram of a track detected by a radar sensor being converted into multiple hypothetical derived Cartesian tracks; a representation of the hypothetical derived Cartesian tracks from multiple sensors and the resulting most plausible fused track; [Diagram 3] 1 is a schematic pre-association graph for multiple radar sensors; [Figure 4] FIG. 4 is a schematic segmentation arrangement taken into account when refining the association graph according to FIG. 3. [Diagram 5] FIG. 2 is a schematic diagram of a data structure for generating a fused track from an average of multiple tracks detected by a radar sensor. [Figure 6] FIG. 1 is a schematic diagram of a poorly plausible fusion track. [Figure 7] FIG. 1 is a schematic diagram of a plausible fusion track. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] Similar items are identified by the same names in all the drawings.

[0013] FIG. 1 shows a block diagram of one possible embodiment of an apparatus 1 for resolving angular ambiguities in a spatially incoherent radar network 2 having multiple radar sensors 2.1 to 2.n.

[0014] Radar sensors 2.1-2.n for vehicle applications can suffer from angle ambiguity. In addition to the usual measurement uncertainties, such radar sensors 2.1-2.n cannot unambiguously detect the angle of incidence of the object's signal. In other words, the radar sensors 2.1-2.n cannot detect the angle of incidence of the object's signal for any angle φ,...,φ N-1It can only be determined that a signal from outside is reaching them, where N is the total number of ambiguities. Using the radar sensors 2.1-2.n it can be determined, for example, which angle ambiguity is most likely based on the illuminated field of view or on internal target tracking. For further processing, the radar sensors 2.1-2.n may output only this most likely angle or other angles that are equally possible.

[0015] In many cases, the most probable angle measurement corresponds to the correct object position. In such cases, tracking and fusion algorithms that ignore the ambiguity of the radar measurements provide good results. If the radar sensors 2.1-2.n incorrectly resolve the angle ambiguity, such algorithms may give very erroneous results. In particular, these may indicate ghost objects, i.e. objects at positions where there are in fact none. These ghost objects may also be long-lived, since the effects of the radar sensors 2.1-2.n incorrectly resolving the angle ambiguity may exist for a long time. In the driving operation of automated, especially highly automated or autonomous vehicles, such ghost objects may have a significant impact on the driving behavior, for example by causing emergency braking for no reason.

[0016] To prevent the recognition of such ghost objects, it is possible to increase the wavelength value for generating objects based on the radar data. However, this would result in a decrease in the reliability and therefore the usefulness of the radar sensors 2.1-2.n. This may also result in additional problems such as delayed recognition or even missing objects.

[0017] Therefore, angular ambiguities in the fusion algorithm and tracking should not be ignored, even if they only appear at a relatively low frequency. As soon as a track is generated, angular ambiguities are recognized and taken into account, especially in the so-called spawning phase, i.e. when the track for the object is initialized, in order to prevent additional ambiguous measurement updates for it from being resolved so that it coincides with the previous resolution. Furthermore, a reliable initialization of the track is necessary in order to be able to deal with emergency situations, such as the appearance of a pedestrian on the road who suddenly enters the field of view of the radar sensors 2.1-2.n, especially when the vehicle equipped with the radar sensors 2.1-2.n is already at a short distance from the pedestrian. Emergency situations can also be caused by obstacles located at a short distance from the vehicle, for example luggage dropped on the road, which, due to their backscattering properties, can only be recognized at a short distance by the radar sensors 2.1-2.n or other sensors.

[0018] To solve these problems, a method for resolving angle ambiguities in a spatially incoherent radar network 2 is implemented using an apparatus 1 .

[0019] The method is adapted to handle and resolve angular ambiguities in radar measurements with object spawns and relies on the use of multiple radar sensors 2.1-2.n, which detect overlapping and / or adjacent detection areas or fields of view around the vehicle on which the sensors are located (hereafter referred to as ego-vehicle). The implementation of the device 1 is based on the goal of avoiding the use of a full "Multiple-Hypothesis Tracking Algorithm", which is more difficult to implement and very computationally intensive.

[0020] The radar sensors 2.1 to 2.n scan the surroundings, where the sensor tracking modules 2.1.1 to 2.n.1 measure tracks T1.1 to T1.m ... Tn.1 to Tn.x of detected objects from the data detected by the radar sensors 2.1 to 2.n and transmit them to the association module 3. The sensor tracking modules 2.1.1 to 2.n.1 generate and manage the tracks T1.1 to T1.m ... Tn.1 to Tn.x independently for each radar sensor 2.1 to 2.n. The tracks T1.1 to T1.m ... Tn.1 to Tn.x are calculated based on the state space of the measurands, in particular the distances of the detected and / or tracked objects to the radar sensors 2.1 to 2.n, the radial velocities v rad , and direction cosine u are used. The direction cosine u specifies the cosine or sine of the incidence angle for each angle definition. By using this special state space, the direction cosine u specifying the incidence angle is used to predict the distance r and radial velocity v to the radar sensors 2.1 to 2.n in the prediction step. rad , the sensor tracking modules 2.1.1 to 2.n.1 do not need to resolve the angle ambiguity.

[0021] Furthermore, the tracks T1.1 to T1.m...Tn.1 to Tn.x are fused into a set of multi-sensor tracks TG1 to TGz by the adjustment submodule 4 of the multi-sensor tracking module.

[0022] The fusion module 5 generates fused tracks S1 to Sy from these multi-sensor track groups TG1 to TGz, respectively.

[0023] However, as shown in more detail in FIG. 2, angular ambiguities must be taken into account when transforming the detected sensor track T1 into derived hypothetical Cartesian tracks T1.1-T1.3.

[0024] In this regard, the left part of FIG. 2 shows a detected sensor track T1, which is transformed, for example by means of equation (1), into a number of hypothetically derived Cartesian tracks T1.1-T1.3, which are shown in the center of FIG.

[0025]

number

[0026] Here, u tr is the direction cosine u of the track s in sensor coordinates where the ambiguity was not resolved. nΔu is the distance between possible resolutions of the ambiguity, and u n is the nth possible solution.

[0027] Each of the three hypotheses corresponds to a possibility for resolving the angle ambiguity.

[0028] In the right part of FIG. 2, hypothetical derived Cartesian tracks T1.1-T1.3, T2.1-T2.3, T3.1-T3.3 and the resulting most plausible fusion track S1 are shown for tracks detected by multiple radar sensors 2.1-2.n.

[0029] The derived Cartesian tracks T1.1-T1.3 are located in a Cartesian coordinate system with coordinates x, y, e.g., an Integrated Driving State Frame, also known as IDS, The derived Cartesian tracks T1.1-T1.3 exist there only from a position point of view.

[0030] Due to the angular ambiguities, there are multiple derived Cartesian tracks T1.1-T1.3, where the number of these tracks T1.1-T1.3 corresponds to the number of ambiguities in the angular measurements. Each derived Cartesian track T1.1-T1.3 is transformed into only one of the multiple hypotheses, such as from the sensor track T1 to the Cartesian tracks T1.1-T1.3.

[0031] One variant of the derived Cartesian tracks T1.1-T1.3 are so-called timestamp-adjusted Cartesian tracks, where "timestamp-adjusted" refers to the fact that for this type of track, timestamps are specified to match in the sensor tracking modules 2.1.1-2.n.1.

[0032] For example, the sensor tracks T1, T2 are transformed into timestamp-adjusted Cartesian tracks T1.1-T1.m...Tn.1-Tn.x with the same timestamp so that these tracks T1.1-T1.m...Tn.1-Tn.x can be merged. Then, t p is the update time of tracks T1.1~T1.m…Tn.1~Tn.x with index p, and t l is a timestamp on the common time axis. Then, t l The state of tracks T1 and T2 with respect to is determined by:

[0033] -Tracks T1 and T2, state ~x p,k and covariance matrix ~P p,k to convert T1.1 to T1.m … Tn.1 to Tn.x into Cartesian tracks with -covariance matrix P p,l State x with p,l To obtain the timestamp t l This includes prediction of the state of

[0034] Tracks T1.1 through T1.m…Tn.1 through Tn.x contain indicator variables ~D p This means that the tracked object is k If the track is recognized by the timestamp, it is set to 1. Otherwise, it is set to 0. For each timestamp-adjusted derived Cartesian track T1.1~T1.m…Tn.1~Tn.x, for each timestamp l, the corresponding variable D p is the previous timestamp t k No~D p In this way, there is a persistent record of whether the tracked object was recognized on a previous occasion.

[0035] For example, fused tracks S1-Sy are created by computing a fused state as a weighted average from the states of the timestamp-adjusted derived Cartesian tracks T1.1-T1.m...Tn.1-Tn.x. In other words, state x p and covariance matrix P p For the timestamp adjusted derived tracks T1.1 to T1.m … Tn.1 to Tn.x with

[0036]

number

[0037] Here, P st is the number of tracks T1.1 to T1.m...Tn.1 to Tn.x that contribute to the fused tracks S1 to Sy, and equation (3) is the covariance matrix of the fused tracks S1 to Sy at time point l.

[0038]

number

[0039] Figure 3 shows an example of a pre-association graph AG. The nodes on the association graph AG are the tracks T1.1-T1.m...Tn.1-Tn.x, and the (weighted) connections represent the pre-associations. Figure 4 shows example segmentation arrangements (a)-(d) considered when refining the association graph AG. Segmentation (a) corresponds to the original pre-associations from the pre-association graph AG. In segmentation arrangements (b), (c) and (d) one or both of the pre-associations are removed.

[0040] The multi-sensor tracking module creates a set of multi-sensor tracks TG1-TGz and assigns to it tracks T1.1-T1.m...Tn.1-Tn.x detected by radar sensors 2.1-2.n possibly indicating the same target object, such that tracks T1.1-T1.m...Tn.1-Tn.x may include tracks T1.1-T1.m...Tn.1-Tn.x from only one radar sensor 2.1-2.n or from multiple radar sensors 2.1-2.n.

[0041] Here, the allocation is done in two steps: In a first step, tracks T1.1-T1.m...Tn.1-Tn.x are allocated between adjacent radar sensors 2.1.-2.n with overlapping detection ranges using a cost matrix and the so-called Munkres algorithm. The cost matrix is ​​based on the plausibility that two tracks T1.1.-T1.m...Tn.1.-Tn.x originate from the same target object. In this first step, an association graph AG is generated whose nodes represent the (weighted) connections of the tracks T1.1-T1.m...Tn.1-Tn.x and pre-associations, as shown in Fig. 3.

[0042] In a second step, connections are removed from the association graph AG. An exemplary case of such removal is when track "A" plausibly originates from the same target object as track "B", and track "B" plausibly originates from the same target object as track "C", but it is not plausible that tracks "A", "B", and "C" all represent the same target object. This case may seem counterintuitive, but can arise due to ambiguous angle measurements.

[0043] A multi-sensor track group TG1-TGz with tracks T1.1.-T1.m…Tn.1.-Tn.x from multiple radar sensors 2.1-2.n provides hypotheses of fused tracks T1.1-T1.m…Tn.1-Tn.x by multiple radar sensors 2.1-2.n. There are multiple hypotheses due to the combination of hypotheses involved in transforming the sensor tracks T1, T2 into derived Cartesian tracks T1.1-T1.m…Tn.1-Tn.x. For example, for a multi-sensor track group TG1-TGz generated from two sensor tracks T1, T2, each track T1, T2 contains three hypothesis transformations to derived Cartesian tracks T1.1-T1.m…Tn.1-Tn.x, and there are a total of 3x3=9 hypothesis fused tracks T1.1-T1.m…Tn.1-Tn.x. This is shown in detail in Figure 5.

[0044] The multi-sensor track flock module uses multiple time-stamp adjusted derived Cartesian tracks T1.1-T1.m...Tn.1-Tn.x, one for each sensor track T1, T2, to create a fused track S1-Sy, where each fused track S1-Sy has a scaled plausibility value, from which the most plausible fused track kS p is selected.

[0045] For example, in a first step, tracks T1.1 to T1.m...Tn.1 to Tn.x are assigned as pre-assigned tracks to a set of multi-sensor tracks TG1 to TGz. In many cases, these pre-associations are maintained. However, there are cases where it is not logical to use them. The purpose of creating the set of multi-sensor tracks TG1 to TGz from the pre-associations includes a heuristic segmentation algorithm. This algorithm first finds all possible segmentation arrangements of the pre-assigned tracks. This allows the pre-associated tracks A, B, and C to be distributed as follows:

[0046] - multi-sensor tracks {A, B} and {C}; -Multi-sensor tracks {A}, {B,C}, -Multi-sensor tracks {A}, {B}, {C} or -Combined multi-sensor tracks {A, B, C}.

[0047] The segmentation algorithm then calculates the validity of each segmentation configuration to select the most valid segmentation configuration. For each segment with a particular segmentation configuration and its index m, the segmentation algorithm constructs a fusion track S1-Sy and calculates its validity W mt , length L mt,m , and the average validity ~W mt,m =W mt,m / L mt,m Next, the validity of the segmentation arrangement is expressed as the following equation (4).

[0048]

number

[0049] where M is the total number of segments in the segmentation arrangement. The most plausible segmentation arrangement is then selected.

[0050] 6 is a schematic diagram of a fusion track S1 with low validity. The original two derived Cartesian tracks T1.1, T1.2 have a very low degree of match. Since the fusion track S1 does not match the derived Cartesian tracks T1.1, T1.2, the algorithm calculates a low validity value.

[0051] 7 is a schematic diagram of a fusion track S1 with high plausibility, where the original two derived Cartesian tracks T1.1, T1.2 have a high degree of agreement, so the fusion track S1 also matches the derived Cartesian tracks T1.1, T1.2, and the algorithm calculates a high plausibility value.

[0052] The plausibility depends on how well each timestamp-adjusted derived Cartesian track T1.1-T1.m...Tn.1-Tn.x matches with the resulting fused track S1-Sy. Furthermore, based on the fused tracks S1-Sy and the sensor model, the likelihood that the tracked target object should have been detected by the associated radar sensor 2.1-2.n is determined. If the actual recognition or non-recognition result matches well with the recognizability, the fusion track S1-Sy has high plausibility. Then, the most plausible fusion track S1-Sy is selected from the multiple hypothetical tracks S1-Sy. p is selected. By computing the various hypotheses of the fused tracks S1 to Sy and their plausibility ranking, the goal of resolving the angle ambiguity is achieved.

[0053] For example, the validity of a fused track is determined using the following heuristic process.

[0054] -If the timestamp-adjusted derived Cartesian tracks T1.1~T1.m…Tn.1~Tn.x do not overlap in time series, the length of the fusion track S1~Sy is 0. In this case, the validity is W mt =0.

[0055] Otherwise, a weight variable is calculated for each derived Cartesian track p and each timestamp l. p For timestamps with a value of 0 (not recognized), the weight w p =1-P D (x l ;v p ), where v p is the ID of the radar sensor 2.1 to 2.n from which the track p is being received, and P D( x;v) is the probability that a target object in state x is recognized by radar sensor 2.1-2.n with IDv. D p,l For timestamps where ,=1 (target is recognized), the weights are determined according to equation (5).

[0056]

number

[0057] where p(x;y,P) is a Gaussian probability density function with mean y and covariance matrix P, and κ is a configurable parameter.

[0058] If the target object is not tracked by the radar sensors 2.1 to 2.n, i.e., there are no tracks T1.1 to T1.m ... Tn.1 to Tn.x from the radar sensors 2.1 to 2.n in the multi-sensor track groups TG1 to TGz, then all time indices are treated as time indices of missed recognition and the weights are w p,l =1-P D( x l ;vp), where p is the index of the missed tracks T1.1-T1.m...Tn.1-Tn.x, and v p is the index of the radar sensor 2.1-2.n that missed the track T1.1-T1.m…Tn.1-Tn.x. These definitions allow us to consistently treat the missed tracks T1.1-T1.m…Tn.1-Tn.x when calculating the fused track validity.

[0059] Next, the validity of the fusion tracks S1 to Sy is given by equation (6).

[0060]

number

[0061] Here, L mt is the number of time steps of the fusion tracks S1 to Sy, and P Sensors is the number of radar sensors 2.1 to 2.n. This value corresponds to the combined number of tracks T1.1 to T1.m...Tn.1 to Tn.x and missed tracks T1.1 to T1.m...Tn.1 to Tn.x.

[0062] The above algorithms are primarily intended for continuous tracking. However, data can be tracked exclusively for short time intervals to receive fused tracks S1-Sy that are used for track initialization or as spawn candidates for the main tracking algorithm. This limited use can simplify the calculation of validity and track fusion. In other words, the situation where sensor tracks T1, T2 represent the same target object for a given time interval, but not for another time interval due to a change in track ID, is not handled. [Explanation of symbols]

[0063] 1 device 2. Radar Network 2.1~2.n Radar Sensor 2.1.1~2.n.1 Sensor Tracking Module 3 Association Module 4. Allocation Submodule 5 Fusion Module (a)~(d) Segmentation arrangement AG Association Graph r removal S1~Sy Tracks S p track T1 Sensor Track T2 Sensor Track TG1~TGz Multi-sensor tracks T1.1~T1.m…Tn.1~Tn.x tracks u direction cosine v rad radial velocity x coordinate y coordinate

Claims

[Claim 1] A method for resolving angular ambiguities in a spatially incoherent radar network (2), comprising: a surrounding area is scanned by a plurality of radar sensors (2.1 to 2.n); for each radar sensor (2.1 to 2.n) individually and independently of each other radar sensor (2.1 to 2.n), a track (T1.1 to T1.m...Tn.1 to Tn.x) of detected objects is generated in a state space; the tracks (T1.1 to T1.m...Tn.1 to Tn.x) of the different radar sensors (2.1 to 2.n) are assigned to one another on condition that there is a plausibility that said tracks originate from the same object, - the mutually assigned tracks (T1.1 to T1.m ... Tn.1 to Tn.x) are fused for different variants to resolve the angle ambiguity; A respective validity measure is assigned to each variant; and the most plausible variant is selected to resolve the angle ambiguity. A method comprising:

Images