METHOD FOR OBJECT CLASSIFICATION USING POLARIMETRICAL RADAR DATA AND SUITABLE DEVICE FOR THIS PURPOSE

Description

[0001] The present invention relates to a method for object classification using polarimetric radar data and a suitable device for this purpose.

[0002] It is generally known to use radars with linearly polarized signals for object classification. The results obtained are not always clear, for example, in recorded radar images, or can be ambiguous with regard to different objects. For instance, DE 10 2013 102 424 A1 describes a polarimetric radar consisting of a transmitting arrangement that emits circularly polarized waves via transmitting antennas and a receiver arrangement that receives the reflected circularly polarized wave components via an antenna arrangement. This arrangement comprises several dual-channel receivers that simultaneously receive left- and right-handed circularly polarized signal components, which are intended for digital beamforming downstream of the antenna arrangement.

[0003] It is therefore an object of the present invention to provide a method for object classification and a suitable device for this purpose, which reduces the disadvantages existing in the prior art. It is also an object of the present invention to improve the accuracy of object classification and, furthermore, to optionally provide data that can be used for a wider range of applications.

[0004] The aforementioned tasks are solved process-technically with the features of claim 1 and device-technically with the features of claim 8.

[0005] This demonstrates that by providing an elliptically or circularly polarized transmission signal directed at the object to be classified, correspondingly different reflection signals are used to generate different radar images, which can then be compared. This method allows for the differentiation of prominent object features, thus enabling improved object classification.

[0006] The method and device according to the application can therefore be used in future radar sensors, which can be used in particular in highly automated and autonomous driving.

[0007] This requires polarimetric radar sensors, which are particularly distinguished by their ability to generate significantly more target information compared to currently used radars with linearly polarized signals. This is because independent radar images can be generated for co- and cross-polarization, and a higher target detection probability exists with circular polarization.

[0008] The method according to the invention involves the evaluation of polarimetric radar data with regard to pattern recognition for the classification of various objects, as well as the detection of so-called "ghost targets". These are caused by multipath propagation, side lobes and periodically recurring main lobes (so-called grating lobes).

[0009] The physical principle applied in the application shows Fig. 1 A circularly or elliptically polarized wave is transmitted, and depending on the structure of the target, a predominantly cross-polarized or a predominantly copolarized wave is received. With an odd number of reflections from the target, the polarization direction reverses, and with an even number of reflections, the same polarization is received. For example, if a left-circular wave is transmitted, the cross-polarization is a right-circular wave, and the copolarization is a left-circular wave. A description of this principle and the construction of a polarimetric radar sensor can be found in [1].

[0010] For this principle to be implemented, a transmitter is required that emits at least one left-circular, right-circular, or elliptical polarization. The receiver must be fully polarimetric. This means that circular, elliptical, and linear polarizations can be received. This can be achieved by receiving the left-circular and right-circular components of the received signal. All polarizations can then be represented by the ratio of the left- and right-circular components. Another way to receive a fully polarimetric signal is by receiving the vertically and horizontally linearly polarized components of the received signal. To represent all polarizations, the magnitude and phase of the vertically and horizontally linearly polarized components of the received signal must be evaluated.

[0011] Fig. 2 This shows the result of a circular polarimetric measurement using a passenger car as an example. Two independent radar images are obtained: a copolar radar image and a crosspolar radar image. The copolar radar image shows even-numbered reflections from the target, primarily double reflections. The crosspolar radar image shows odd-numbered reflections from the target, primarily single reflections. In both cases, local maxima of the reflected signal amplitudes are represented by circles whose diameter is proportional to their amplitude.

[0012] Fig. 3a und Fig. 3b show a polarimetric radar image that shows both the copolar and crosspolar local maxima from Fig. 2 This includes, however, the ratio of the magnitude between the co-polarized and cross-polarized signal components (hereinafter referred to as "ratio") is now additionally represented for each local maxima in the form of different symbols. This creates a polarimetric pattern in different areas of the car, which can be used to classify objects.

[0013] When pattern recognition for different areas of the targets, the following properties of the local maxima are evaluated: Number of copolar maxima, including the possibility of including specific signal-to-noise ratios (distance to noise level); Number of crosspolar maxima, including the possibility of including specific signal-to-noise ratios (distance to noise level); Average magnitude ratio between co- and cross-polarization; Maximum magnitude ratio between co- and cross-polarization; Minimum magnitude ratio between co- and cross-polarization; Phase ratios between co- and cross-polarization; Particularly characteristic properties

[0014] The latter is, for example, the first reflection from the car in the area of ​​the front license plate. This has a ratio of less than -20 dB.

[0015] The pattern classification distinguishes between different object types such as cars, pedestrians, cyclists, trucks and motorcyclists, and road construction objectives such as gullies, barriers, guardrails, bridges and tunnels.

[0016] Fig. 4 The image shows the car measured head-on and at an angle of -20°. At an angle of -20°, the following change can be observed compared to the car measured head-on: The characteristic reflection at the front with a ratio of less than -20 dB shifts to the right side of the car. Reflections with ratios between 10 dB and 15 dB occur in the front area, as well as a reflection with a ratio between 15 dB and 20 dB. (This is due to the stronger copolar properties. These arise because the angled vehicle results in more double reflections in the front area, particularly around the grille.) This leads to a change in the polarimetric pattern, especially in the front, around the steering wheel, and in the rear of the vehicle.

[0017] Based on these properties, it is possible to determine the angle at which the vehicle is located relative to the sensor.

[0018] Fig. 5 This shows areas that are particularly important for pattern recognition and classification of a frontally measured passenger car. These are: Front area, front wheel arches, steering wheel area, exterior mirrors

[0019] The characteristic feature here is accordingly Fig. 3 , a strong reflection in the area of ​​the front license plate with a very low ratio (strong single reflection).

[0020] Fig. 6 shows areas that are of particular importance in pattern recognition and classification of a slant-detected car: Front area, the wheel arches aligned with the sensor, the exterior mirror aligned with the sensor, the front door gap aligned with the sensor, the rear corner of the vehicle aligned with the sensor

[0021] Particularly characteristic here is the detection of the vehicle's contour as an L-shape, the exact position detection of the wheel arches, as well as the occurrence of a relatively large number of signals with a high ratio (double reflections) compared to the other measurement positions.

[0022] Fig. 7 shows areas that are of particular importance in pattern recognition and classification of a cross-detected passenger car: the transverse area aligned with the sensor, the wheel arches aligned with the sensor, the vehicle corners aligned with the sensor, the front door aligned with the sensor

[0023] Particularly characteristic is the detection of strong reflections with a very low ratio (strong single reflections) in the area of ​​the front door.

[0024] Fig. 8 shows areas that are of particular importance in pattern recognition and classification of a car detected from behind: Rear vehicle area, the vehicle corners aligned with the sensor, the front door aligned with the sensor

[0025] Particularly characteristic is the detection of a strong reflection with a very low ratio (strong single reflections) on the outer contour of the rear, as well as a polarimetric pattern originating from the interior of the car.

[0026] Typical polarimetric patterns with the described properties of object classes or object subclasses are always assigned to different angle and distance ranges and serve as the basis for a classification algorithm.

[0027] Furthermore, circular polarization offers advantages in detecting "ghost targets" caused by multipath propagation, side lobes, or interfering recurring main lobes. The latter two are particularly pronounced when detecting strong targets at steep angles.

[0028] Fig. 9 This illustrates the situation with multipath propagation caused by an additional reflection from an object. This creates a mirrored target located at a different angle. However, the additional reflection also causes a rotation of the polarization properties, so that analyzing the altered polarization pattern allows for the identification of multipath propagation.

[0029] Fig. 10 shows the same situation as in Fig. 9 This refers exclusively to targets with a relative velocity to the radar sensor and the environment. If, for targets within a range and angle gate, a local maximum exhibits different velocities in the copolar and crosspolar signals, then it is a "ghost target." These can be caused by multipath propagation, as well as by sidelobes or recurring mainlobes.

[0030] In radar sensors that emit circular polarimetric or elliptical waves, the reflected signals can be decomposed into left- and right-handed components. This yields a polarimetric pattern that can be used for object classification. To receive both the left- and right-handed components, a logical approach is to provide corresponding receiver channels for each polarization. However, this introduces a significant disadvantage: compared to a linear radar system, twice the number of receiver channels is required for the same angular resolution.

[0031] A solution without this disadvantage is found in the method according to the invention. Left- and right-handed waves are transmitted alternately one after the other, and only one polarization direction is received. For example, when left-handed signals are received, the copolar signal components are received when the left-handed wave is transmitted, and the crosspolar signal components are received when the right-handed wave is transmitted. Fig. 11 illustrates the method according to the invention.

[0032] When using multiple transmitters, a logical approach is to operate them sequentially and account for the time-shifted received signals in the signal analysis. However, the long transmission duration results in a significant disadvantage. While very long observation periods yield excellent velocity resolution, high velocities can no longer be unambiguously determined.

[0033] One solution is offered by the method according to the invention. Several transmitters are operated simultaneously and each is phase-coded individually, with the transmitters always having the same polarization. For example, first all left-handed polarized signals are transmitted simultaneously in phase encoding, and then, after a time delay, all right-handed polarized signals are transmitted in phase encoding. In general, the phase encoding can have different durations. Fig. 12 illustrates the method according to the invention.

[0034] When measuring object heights at large distances, there is a well-known method that uses radar devices emitting linearly polarized signals. When measuring objects on the road, signal superposition occurs, caused by different propagation paths. Direct detection is superimposed with multipath propagation consisting of additional reflection from the road surface and two so-called circular paths. A circular path means that the outbound and return paths are different. In the first circular path case, the outbound path is the direct path, and the return path includes road reflection. In the second circular path case, the outbound path includes road reflection, and the return path is the direct path. The superposition of these different propagation paths results in a received signal consisting of the superposition of the reflected signals from the individual propagation paths.Thus, the received signal exhibits an object-dependent characteristic curve over distance, which is determined by the partly constructive and partly destructive superposition of the various signals, depending on the object's distance. If objects with a relative velocity to the sensor are tracked over distance using a tracker, the object's height can be determined by detecting at least two characteristic features, such as two minima of the received signal. However, there is a significant disadvantage in the linear case. Due to the superposition of four different propagation paths, a characteristic curve results, the minima of which are locally very pronounced.Since, at long distances, the reflected signal amplitude generally has a small margin above the noise level, in the linear case, at a certain distance the target can no longer be detected because the received signal is below the noise level. Fig. 13 outlines the propagation paths that occur in the known method, which refers to linearly polarized transmitted signals.

[0035] In the inventive method for determining object heights, circularly or elliptically polarized signals are transmitted. Either a left-handed or a right-handed polarized signal is transmitted, but only the cross-polarized received signal is evaluated. The cross-polarized signal, or copolar signal, always denotes the polarization direction relative to the transmitted signal. If the transmitted signal is, for example, a left-handed wave, then the cross-polarized received signal is right-handed and the copolarized received signal is left-handed. When evaluating only the cross-polarized received signal according to the inventive method, in contrast to the known approach with linearly polarized signals, only two propagation paths result, the reflected signals of which superimpose at the receiver. The propagation paths consist of direct detection and multipath propagation, which includes additional reflection from the road surface.In these two propagation paths, there is an odd number of reflections, and thus the received signal appears in the cross-polarized receiving channel. The circular paths that occur in the known method with linearly polarized signals no longer exist in the method according to the invention, since these signals occur in the copolar receiving path because the number of reflections in the propagation path is even. The strongly pronounced minima that occur locally in the known method, which prevent object detection at certain distances, no longer occur in the method according to the invention, and the objects can be detected at all distances. The height can be determined by evaluating typical features such as two local minima at specific distances. Fig. 14 outlines the propagation paths in the inventive method for determining the object height. Fig. 15 The graph shows typical characteristic curves over distance for both the known method and the method according to the invention. It is worth noting that the respective local minima are generally located above the noise level. In contrast, the local minima in the known method are often below the noise level and therefore difficult to identify. Fig. 16 shows the formulaic relationships for determining the object height.

[0036] Another application of the method and device according to the application is to determine the coefficient of friction of road surfaces, preferably by means of radar sensors, when, for example, additional reflections on the road surface are measured.

[0037] In particular, highly automated or autonomous driving requires predictive measurement of the road surface's coefficient of friction. The measurement results allow for the determination of speeds at which, for example, a curve can be safely negotiated without the risk of the vehicle drifting.

[0038] Measuring the coefficient of friction requires a circularly or elliptically polarized radar sensor, mounted facing forward on the vehicle, which detects the entire surface of the road or pavement in front of the vehicle. This setup is in Fig. 17 shown.

[0039] The following procedure is required to determine the coefficient of friction: 1. Calculation of the local maxima for the co- and cross-polarized reflected signals (left- and right-handed signals). 2. Selection of an area containing a specific road surface. 3. Analysis of all local maxima within this area with respect to the following properties: Amplitude ratio between co- and cross-polarized signal components; Phase difference between co- and cross-polarized signal components; Amplitude strength and polarization of the local maxima.

[0040] To illustrate this, it is helpful to plot this parameter in a graph. Fig. 18 The results for a road with a tar surface exhibiting low roughness are shown as an example and in Fig. 19 As an example, the results for a gravel road with a naturally high roughness are shown. The y-axis shows the ratio of the amplitudes of the copolar and crosspolar signal components, and the x-axis shows their phase difference. The crosspolar local maxima are represented as circles, and the copolar local maxima as rectangles. The size of the circles and rectangles increases with the signal amplitude.

[0041] The coefficient of friction of road surfaces can now be determined using the following properties: Cluster formation, dispersion, cluster size, mean phase difference positions of the clusters, mean amplitude ratios of the clusters, standard deviation of the phase difference positions of the clusters, standard deviation of the amplitude ratios of the clusters, number of local maxima, signal amplitude of the local maxima

[0042] According to Fig. 18 and Fig. 19Road surfaces with a low coefficient of friction exhibit the following properties compared to road surfaces with a high coefficient of friction: Small cluster size; lower standard deviation of phase differences and amplitude ratios; lower signal amplitudes; fewer local maxima

[0043] At low coefficients of friction, these properties are caused by a similar shape and orientation of the backscatter points relative to the sensor. At very high coefficients of friction, backscatter points of varying shapes and orientations relative to the sensor are observed.

[0044] Furthermore, the phase relationship of the clusters allows for a more precise analysis of different surfaces, such as snow, ice, or leaf-covered areas. When water is present on the road, all radar signals are reflected away, and these can be identified by means of signal-free areas in the radar image (i.e., areas where a radar signal is absent), thus enabling the detection of aquaplaning hazards.

[0045] Advantageous further developments are the subject of dependent claims.

Claims

1. A method for object classification which comprises the following steps: a. providing an elliptically or circularly polarized transmission signal which is transmitted to the object to be classified, b. generating a first radar image from the co-polarized reflection signal and generating a second radar image from the cross-polarized reflection signal, and c. comparing the first radar image with the second radar image as a basis for the object classification, characterized in that, in addition to the object classification, the height of objects is determined by detecting an object located on a road at different distances and by evaluating the minima solely of the cross-polar reception signal component related to the transmission signal in dependence on the object distance.

2. A method according to claim 1, wherein the generation of the first and second radar image takes place from left-circular signal components and / or right-circular signal components.

3. A method according to claim 1, wherein the generation of the first and second radar image takes place from linearly horizontal and linearly vertical signal components.

4. A method according to any one of the claims 1 to 3, wherein the generation of the first and second radar image takes place simultaneously, preferably by means of the transmission signal, and / or preferably the comparison takes place using the signal properties of the local maxima of the reflection signals and / or preferably the comparison takes place using the signal properties of individual target regions of the object and / or preferably the signal properties have at least one of the following criteria: - number of co-polar local maxima, - number of cross-polar local maxima, - magnitude ratio between co-polarization and cross-polarization, preferably average magnitude ratio thereof, - maximum magnitude ratio between co-polarization and cross-polarization, - minimum magnitude ratio between co-polarization and cross-polarization, - phase relationship between co-polarization and cross-polarization, - location of local maxima with a high or low magnitude ratio between co-polarization and cross-polarization, - speed evaluation.

5. A method according to any one of the claims 1 to 4, wherein similarity patterns are created for an object classification and / or preferably further sub-object classes are provided that are divided, at least in the case of a radar sensor transmitting the transmission signal, into - a distance from the radar sensor, - an angular alignment with respect to the radar sensor, - an object alignment with respect to the radar sensor, - a relative speed with respect to the radar sensor, and / or preferably the object classification relates to the position and orientation recognition of the object, preferably a passenger car, and / or preferably at least the following areas of a passenger car are evaluated in a frontal object determination, e.g. the front area, front wheel housings, steering wheel area and / or exterior mirrors.

6. A method according to any one of the claims 1 to 5, wherein a left-circularly polarized wave and then a right-circularly polarized wave are transmitted offset in time or a left-rotating elliptical wave and then a right-rotating elliptical wave are transmitted offset in time by at least one transmission antenna and, of the signals reflected back, only the left-rotating or the right-rotating signal component is evaluated.

7. A method according to any one of the claims 1 to 6, wherein, when using a plurality of transmitters, the respective identically polarized waves are transmitted simultaneously and phase-coded so that there is a transmission sequence which consists of a simultaneous use of right-rotating polarized transmission signals and, offset in time, i.e. preceding or following, of a simultaneous use of left-rotating polarized transmission signals.

8. An apparatus for defining an object classification in particular using the method according to any one of the claims 1 to 7, wherein a. means for providing an elliptically or circularly polarized transmission signal which is transmitted to the object to be classified, b. means for generating a first radar image from the co-polarized reflection signal and for generating a second radar image from the cross-polarized reflection signal, and c. means for comparing the first radar image with the second radar image are provided as a basis for the object classification, characterized in that, in addition to the object classification, the height of objects is determined by detecting an object located on a road at different distances and by evaluating the minima solely of the cross-polar reception signal component related to the transmission signal in dependence on the object distance.

9. An apparatus according to claim 8, wherein means for generating the first and second radar image from left-circular or right-circular signal components are provided.

10. An apparatus according to claim 8, wherein means for generating the first and second radar image from linearly horizontal and linearly vertical signal components are provided.

11. An apparatus according to any one of claims 8 to 10, in which the generation of the first and second radar image takes place simultaneously, preferably by means of the transmission signal.

12. An apparatus according to any one of the claims 8 to 11, in which the comparison takes place using the signal properties of the local maxima of the reflection signals.

13. An apparatus according to claim 12, in which the comparison takes place using the signal properties of individual target regions of the object and / or preferably a left-circularly polarized wave and then a right-circularly polarized wave are transmitted offset in time or a left-rotating elliptical wave and then a right-rotating elliptical wave are transmitted offset in time by at least one transmission antenna and, of the signals reflected back, only the left-rotating or the right-rotating signal component is evaluated and / or, preferably when using a plurality of transmitters, the respective identically polarized waves are transmitted simultaneously and phase-coded so that there is a transmission sequence which consists of a simultaneous use of right-rotating polarized transmission signals and, offset in time, i.e. preceding or following, of a simultaneous use of left-rotating polarized transmission signals.

14. An apparatus according to any one of the claims of 8 to 13, wherein, in addition to the object classification, the friction coefficient of a road surface is determined by using a specific area of the road surface for analysis and by evaluating its local maxima at least with respect to at least one of the following properties: - cluster formation, scattering, - expansion of the clusters, - mean value of the phase difference positions of the clusters, - mean value of the amplitude magnitude ratios of the clusters, - standard deviation of the phase difference positions of the clusters, - standard deviation of the amplitude magnitude ratios of the clusters, - number of local maxima, - signal amplitude of the local maxima.