DE102018000517A1 - Method for radar-based measurement and / or classification of objects in a vehicle environment - Google Patents

Abstract

The invention relates to a method for radar-based measurement and / or classification of objects (2) in a vehicle environment, wherein the vehicle surroundings are detected by means of at least one radar sensor (3) arranged on a vehicle (1) and in a determination and / or classification of a height (H) of an object (2) based on an evaluation of a shift of a Doppler frequency between a radar sensor (3) emitted and a reflected from the object (2) radar signal Doppler information are generated. According to the invention, the determination and / or classification of the height (h) of the object (2) is carried out by means of at least one trained neural network, with training of the neural network based on simulation data generated from the Doppler information.

Description

The invention relates to a method for radar-based measurement and / or classification of objects in a vehicle environment according to the preamble of claim 1.

From the DE 10 2015 009 382 A1 a method is known for radar-based determination of a height of an object in a vehicle environment, wherein the vehicle surroundings are detected by means of at least one radar sensor arranged on a vehicle. A height of the object is determined by means of an evaluation of a shift of a Doppler frequency between a radar sensor emitted by the radar sensor and a radar signal reflected by the object, wherein the height in the evaluation of a determined taking into account a approach speed of the vehicle to the object temporal Doppler frequency course based on a comparison with deposited Pattern Doppler frequency curves or a ratio of a radial velocity component of the approach speed of the vehicle to the object and a longitudinal speed component of the approach speed is determined in consideration of a distance of the vehicle to the object.

The invention is based on the object to provide a comparison with the prior art improved method for radar-based measurement and / or classification of objects in a vehicle environment.

The object is achieved by a method having the features specified in claim 1.

Advantageous embodiments of the invention are the subject of the dependent claims.

In the method for radar-based measurement and / or classification of objects in a vehicle environment, the vehicle environment is detected by means of at least one radar sensor arranged on a vehicle, and in a determination and / or classification of a height of an object, an evaluation of a shift of a Doppler frequency between one of Radar sensor emitted and a reflected from the object radar signal Doppler information generated.

According to the invention, the determination and / or classification of the height of the object is carried out by means of at least one trained neural network, with training of the neural network based on simulation data generated from the Doppler information.

The method enables the measurement and / or classification of objects in the vehicle environment, in particular the determination of the height of objects, in a particularly reliable and simple manner. In this case, a number of misdetections of over or under traversable objects is significantly reduced, whereby an increase in reliability of driver assistance systems for autonomous or semi-autonomous longitudinal and / or lateral control of a vehicle, such as parking assistance systems, is achieved because of a lane elevated and non- raised objects can be distinguished from each other particularly reliable. Thus, faulty controls of the vehicle, for example, to avoid collisions with non-raised objects or objects with low height, such as curbs avoided. Furthermore, a robustness of radar-based driver assistance systems, in particular an increase in the robustness of a localization of the vehicle based on the radar data, is increased.

Embodiments of the invention are explained in more detail below with reference to drawings.

• 1 schematisch eine Draufsicht eines Fahrzeugs, eines vor dem Fahrzeug befindlichen Objekts und geometrische Verhältnisse zwischen einem Radarsensor des Fahrzeugs und dem Objekt und
• 2 schematisch eine Seitenansicht des Fahrzeugs und des Objekts gemäß 1.

Showing:

• 1 schematically a plan view of a vehicle, an object located in front of the vehicle and geometric relationships between a radar sensor of the vehicle and the object and
• 2 schematically a side view of the vehicle and the object according to 1 ,

Corresponding parts are provided in all figures with the same reference numerals.

In 1 are a top view of a vehicle 1 , one in front of the vehicle 1 located object 2 and geometric relationships between a radar sensor 3 of the vehicle 1 and the object 2 shown. 2 shows a side view of the vehicle 1 and the object 2 ,

The vehicle 1 is designed in particular for an autonomous or semi-autonomous driving operation and comprises at least one driver assistance system for carrying out a so-called automatic valet parking.

The vehicle 1 is in a global coordinate system with the axes x . y and the radar sensor 3 described in a sensor coordinate system with the axes x ', y'.

The vehicle 1 moves at a speed v which consists of a velocity component v _x in the direction of the axis x and a speed component v _y composed. A size of the velocity components v _x . v _y is of a yaw rate w of the vehicle 1 dependent.

In the present embodiment, a radar sensor 3 which is concerned with a sensor speed v _s moved in a scene and a stationary object 2 detected. For simplicity, it is assumed that the scene is a two-dimensional plane, so that an elevation angle ε of the observed object equals 0 with respect to a line of sight of the radar sensor 3 is. It is also assumed that the line of sight of the radar sensor 3 is a vector corresponding to the axis x 'of the local sensor _coordinate system, such that a measured radial velocity v _{r2d of} the object 2 at the receiver of the radar sensor 3 is given by $$-v_{r2d}=-v_{x\text{'}}\text{cos}\left({\theta }_S\right)+v_{y\text{'}}\text{sin}\left({\theta }_S\right),$$

In this case, θ _S denotes a sensor azimuth angle of the object 2 in the local sensor coordinate system and v _{x '} and v _y' denote sensor velocity components along the axes x ', y'. To validate the radial velocity formulation for any sensor orientations, a transformation of the sensor velocity components v _{x '} and v _y' from the sensor coordinate system into a global coordinate system is necessary. To do this, the local sensor axes become an origin around a sensor mounting angle β turned to global axes x . y of the global coordinate system. This results in the radial velocity v _{r2d of} the object 2 : $$-v_{r2d}=-v_x\text{cos}\left(\theta \right)+v_y\text{sin}\left(\theta \right),$$

und ein Azimutwinkel θ des Objekts 2 zu $$\theta ={\theta }_S+\beta .$$

In this case, the sensor velocity components v _{x '} and v _y' in the global coordinate system result $$\left[\begin{array}{c}{v}_x\\ v_y\end{array}\right]=\left[\begin{array}{cc}\text{cos}\left(\beta \right)& \text{sin}\left(\beta \right)\\ -\text{sin}\left(\beta \right)& \text{cos}\left(\beta \right)\end{array}\right]\left[\begin{array}{c}{v}_{x\text{'}}\\ v_{y\text{'}}\end{array}\right]$$

and an azimuth angle θ of the object 2 to $$\theta ={\theta }_S+\beta ,$$

$$v_{y\text{'}}=wx\text{'}+v_y.$$

For the moving vehicle 1 mounted radar sensor 3 can be through the speed v and the yaw rate w described self-motion of the vehicle 1 together with a sensor mounting position in the global coordinate system described by the coordinates x '= 0 and y' = 0. The origin of the global coordinate system is located in the middle of a vehicle's rear axle in such a way that the axis x parallel and the axis y is perpendicular to the direction of movement. The sensor speed components v _{x '} , v _y' are due to the rotation of the vehicle 1 with the yaw rate w around the origin of the coordinates and because of the speed v of the vehicle 1 to $$v_{x\text{'}}=-wy\text{'}+v_x\text{and}$$

$$v_{y\text{'}}=wx\text{'}+v_y,$$

From equations (5) and (6), equation (2) yields: $$-v_{r2d}=\left[-y\text{'}\text{cos}\left(\theta \right)+x\text{'}\text{sin}\left(\theta \right)\right]w+\text{cos}\left(\theta \right)v_x+\text{sin}\left(\theta \right)v_y,$$

In a three-dimensional scenario contains the measured radial velocity v _r a Doppler shift component which depends on the elevation angle ε of the object 2 , That is, equation (7) is extended to: $$\begin{array}{cc}-v_r& =-v_{r2d}\text{cos}\left(\epsilon \right)\\ & =\left[\left(-y\text{'}\text{cos}\left(\theta \right)+x\text{'}\text{sin}\left(\theta \right)\right)w+\text{cos}\left(\theta \right)v_x+\text{sin}\left(\theta \right)v_y\right]\text{cos}\left(\epsilon \right),\end{array}$$

If there is no wheel drift, the speed becomes v on the vehicle rear axle to the speed component v _x reduced. This results in: $$-v_r=\left[\left(-y\text{'}\text{cos}\left(\theta \right)+x\text{'}\text{sin}\left(\theta \right)\right)w+\text{cos}\left(\theta \right)v_x\right]\text{cos}\left(\epsilon \right),$$

wobei R einen radialen Abstand vom Radarsensor 3 zum Objekt 2 und h_s eine Sensorhöhe beschreiben. Der Elevationswinkel ε kann aus der Gleichung (9) als Funktion von der Radialgeschwindigkeit v_r des Objekts 2, der Sensorbefestigungsposition mit den Koordinaten x' =0 und y'= 0, dem Azimutwinkel θ des Objekts 2 und der Eigenbewegung des Fahrzeugs 1, bestehend aus der Geschwindigkeitskomponente v_x und der Gierrate w ermittelt werden.A height h _t of the object 2 is given by $$H_t=H_{s}R\text{sin}\left(\epsilon \right).$$

where R is a radial distance from the radar sensor 3 to the object 2 and h _s describe a sensor height. The elevation angle ε can be calculated from the equation (9) as a function of the radial velocity v _r of the object 2 , the sensor mounting position with the coordinates x '= 0 and y' = 0, the azimuth angle θ of the object 2 and the proper motion of the vehicle 1 consisting of the velocity component v _x and the yaw rate w be determined.

Thus results for the elevation angle ε : $$\epsilon =\frac{-v_r}{=\left[\left(-y\text{'}\text{cos}\left(\theta \right)+x\text{'}\text{sin}\left(\theta \right)\right)w+\text{cos}\left(\theta \right)v_x\right]},$$

Provided that accurate movement information of the vehicle 1 are available, the height can be h _t of the object 2 be determined by the measured information about the azimuth angle θ of the object 2 exploited along with the received Doppler information to the elevation angle ε to calculate exactly. The azimuth angle θ of the object 2 is digitally beamformed to several horizontal antennas of the radar sensor 3 certainly. Is the elevation angle ε of the object 2 once calculated, the height can be h _t of the object 2 from the elevation angle ε of the object 2 and the radial distance R from the radar sensor 3 to the object 2 , as indicated in equation (10).

To determine the height h _t of the object 2 By means of a so-called "Doppler Beam Sharpening Algorithm", the calculated Doppler shift of the received signal and azimuth angle information together with the intrinsic motion information of the vehicle 1 combined as described above such that the elevation angle ε of the object 2 is determined. The azimuth angle θ of the object 2 is using digital beamforming to several horizontal antennas of the radar sensor 3 certainly.

For this purpose, objects are first 2 detected and extracted. To accomplish this, a coarse 3D Fast Fourier Transform is performed on the received pulse signal from the radar sensor 3 performed, providing output range, output angle and output Doppler estimates.

Subsequently, a 2D cell average constant false alarm rate (= CA-CFAR) is applied to an output of the 3D Fast Fourier Transform for peak detection. Then the detected peak values, also referred to as peaks, are merged and the objects 2 extracted using a Density-Based Spatial Clustering of Applications with Noise (DBSCAN). Subsequently, with one out "J. Li and P. Stoica: Efficient mixed-spectrum estimation with applications to target feature extraction; Signal Processing, IEEE Transactions on, vol. 44, no. 2, pp. 281-295, 1996 " known 3D recursive Fourier transform "RELAX" calculated a finer 3D spectral estimate of the pulse signal. Finally, the height is determined h _t of the object 2 based on 3D parameters obtained from RELAX and the geometry model of the Doppler Beam Sharpening Algorithm.

• Schritt 1: Grobe 3D-Paramterschätzung;
• Schritt 2: Peakerkennung und Objektextraktion;
• Schritt 3: Hochauflösende 3D-Paramterschätzung;
• Schritt 4: Objekthöhenermittlung.

The determination of the height h _t of the object 2 Using the "Doppler Beam Sharpening Algorithm" can thus be summarized in four main steps:

• step 1 : Rough 3D parameter estimation;
• step 2 : Peak detection and object extraction;
• step 3 : High-resolution 3D parameter estimation;
• Step 4: object height determination.

In order to realize an accurate object height determination by means of the "Doppler Beam Sharpening Algorithm", the 3D object parameters must be estimated as accurately as possible. For this, a three-dimensional RELAX algorithm is implemented to allow high resolution and accurate angle and Doppler estimation. After the object extraction, a high-resolution three-dimensional spectral power density is generated by means of 3D-RELAX from the pulse signal according to equation (7).

RELAX is a high resolution parametric algorithm based on a nonlinear regression model to estimate the frequencies present in a signal and their respective amplitudes. To get the 3D parameter of the extracted K objects 2 For example, RELAX assumes that the received radar signal S _IF (n, l, m) can be modeled as the sum of several weighted complex exponentials plus noise. Thus, the received radar signal S _IF (n, l, m) can be described by the following model: $$y_{n.l.m}={\displaystyle \underset{k=1}{\overset{K}{\Sigma }}{\alpha }_k{e}^{j2\pi f_{k}n}{e}^{j2\pi f{\text{'}}_{k}l}{e}^{j2\pi f\text{'}{\text{'}}_{k}m}+e_{n.l.m,}}$$

α_k können durch rekursives Minimieren der folgenden nichtlinearen Kleinste-Quadrate-Kostenfunktion ermittelt werden: $$C_l{\left(\left\{f_i,f{\text{'}}_i,f\text{'}{\text{'}}_i,{\alpha }_i\right\}\right)}_{i=0}^K={\left|\left|y-{\displaystyle \sum _{k=1}^K{\alpha }_{k}W\left(f_k\right)}\right|\right|}^2,$$

wobei $\left||\dots |\right|$

der euklidische Betrag, y das dreidimensionale NxLxM Pulssignal in Gleichung (7) und W(f_k) = [W_n,l,m] ein Tensor dritter Ordnung ist, so dass $$W_{n,l,m}=w_n\left(f_k\right)w_n\left(f{\text{'}}_k\right)w_n\left(f\text{'}{\text{'}}_k\right)$$

ist, wobei w_n(f_k)w_n(f'_k)w_n(f"_k) gegeben sind durch $$\begin{array}{c}{w}_n\left(f_k\right)={\left[1e^{j2\pi f_k}\dots e^{j2\pi f_k\left(N-1\right)}\right]}^T,\\ w_n\left(f{\text{'}}_k\right)={\left[1e^{j2\pi f_k^{\text{'}}}\dots e^{j2\pi f_k^{\text{'}}\left(L-1\right)}\right]}^T,\\ w_n\left(f\text{'}{\text{'}}_k\right)={\left[1e^{j2\pi f_k^{\text{'}\text{'}}}\dots e^{j2\pi f_k^{\text{'}\text{'}}\left(M-1\right)}\right]}^T.\end{array}$$

Where n = 0; 1; ...; N - 1, l = 0; 1; ...; L - 1 and m = 0; 1; ...; M - 1, where L denotes a number of received channels, N a number of samples per chirp, and M a number of ramps per frame. The variable a _K denotes the unknown complex amplitude and f _k , f ' _k , f " _k denote the unknown range, angular and Doppler frequencies of the k th harmonic, for k = 1; 2; ...; K. e _{n, l, m} describes the noise. The sinusoidal parameters $\widehat{f_k}.\widehat{f{\text{'}}_k.}\widehat{f\text{'}{\text{'}}_k}.$

α _k can be determined by recursively minimizing the following nonlinear least squares cost function: $$C_l{\left(\left\{f_i.f{\text{'}}_i.f\text{'}{\text{'}}_i.{\alpha }_i\right\}\right)}_{i=0}^K={\left|\left|y-{\displaystyle \underset{k=1}{\overset{K}{\Sigma }}{\alpha }_{k}W\left(f_k\right)}\right|\right|}^2.$$

in which $\left||...|\right|$

the Euclidean amount, y the three-dimensional NxLxM pulse signal in equation (7) and W (f _k) = [W _{n, l, m]} is a tensor of the third order, so that $$W_{n.l.m}=w_n\left(f_k\right)w_n\left(f{\text{'}}_k\right)w_n\left(f\text{'}{\text{'}}_k\right)$$

where w _n (f _k ) w _n (f ' _k ) w _n (f " _k ) are given by $$\begin{array}{c}{w}_n\left(f_k\right)={\left[1e^{j2\pi f_k}...e^{j2\pi f_k\left(N-1\right)}\right]}^T.\\ w_n\left(f{\text{'}}_k\right)={\left[1e^{j2\pi f_k^{\text{'}}}...e^{j2\pi f_k^{\text{'}}\left(L-1\right)}\right]}^T.\\ w_n\left(f\text{'}{\text{'}}_k\right)={\left[1e^{j2\pi f_k^{\text{'}\text{'}}}...e^{j2\pi f_k^{\text{'}\text{'}}\left(M-1\right)}\right]}^T,\end{array}$$

wobei ${\left\{{\widehat{f}}_i,{\widehat{\alpha }}_i\right\}}_{i=\mathrm{1,}i\ne k}^K$

angenommen wird und die Kostenfunktion gemäß Gleichung (13) bezüglich f_k und α_k minimiert wird. Somit ergibt sich: $$\left({\widehat{f}}_k,{\widehat{f}}_k^{\text{'}},{\widehat{f}}_k^{\text{'}\text{'}}\right)=\underset{f_k,f_k^{\text{'}},f_k^{\text{'}\text{'}}}{{\displaystyle \text{max}}}{\left|<W\ast \left(f_k\right),y_k>\right|}^2$$

und $$\widehat{{\alpha }_k}=\frac{1}{NLM}<W\ast \left(f_k\right),y_k>.$$

Let y _{k be} given by $$y_K=y-{\displaystyle \underset{i=0i\ne k}{\overset{K}{\Sigma }}{\widehat{\alpha }}_{i}W\left({\widehat{f}}_i\right).}$$

in which ${\left\{{\widehat{f}}_i.{\widehat{\alpha }}_i\right\}}_{i=\mathrm{1,}i\ne k}^K$

is assumed and the cost function according to equation (13) with respect to f _k and α _{k is} minimized. This results in: $$\left({\widehat{f}}_k.{\widehat{f}}_k^{\text{'}}.{\widehat{f}}_k^{\text{'}\text{'}}\right)=\underset{f_k.f_k^{\text{'}}.f_k^{\text{'}\text{'}}}{{\displaystyle \text{Max}}}{\left|<W*\left(f_k\right).y_k>\right|}^2$$

and $$\widehat{{\alpha }_k}=\frac{1}{NLM}<W*\left(f_k\right).y_k>,$$

kann als Ort des dominanten Peaks des Periodogramms bzw. Wellenschaubilds ermittelt werden, wobei das Periodogramm effizient mittels der 3D-Fast-Fourier-Transformation des Datenkubus y_k unter Verwendung einer Auffüllung mit Nullen berechnet werden kann.Where <...> is the sum of all elemental products. The estimate of $\left({\widehat{f}}_k.{\widehat{f}}_k^{\text{'}}.{\widehat{f}}_k^{\text{'}\text{'}}\right)$

can be determined as the location of the dominant peak of the periodogram, and the periodogram can be calculated efficiently using the 3D Fast Fourier Transform of the data cube y _k using a padding with zeros.

den Azimutwinkel θ des Objekts 2 mit $$\theta =\text{arcsin}\left(\frac{f_k^{\text{'}}\lambda }{d}\right)$$

und die Radialgeschwindigkeit v_r mit $$v_r=f_k^{\text{'}\text{'}}\frac{c}{2f_{c}T}$$

aus $\left({\widehat{f}}_k,{\widehat{f}}_k^{\text{'}},{\widehat{f}}_k^{\text{'}\text{'}}\right)$

zu extrahieren, wobei c die Lichtgeschwindigkeit, f_s die Samplingfrequenz und B die Radarbandbreite beschreibt.Thus it is possible the distance R from the radar sensor 3 to the object 2 With $$R=\frac{f_k{f}_{s}cT}{2B}.$$

the azimuth angle θ of the object 2 With $$\theta =\text{arcsin}\left(\frac{f_k^{\text{'}}\lambda }{d}\right)$$

and the radial speed v _r With $$v_r=f_k^{\text{'}\text{'}}\frac{c}{2f_{c}T}$$

out $\left({\widehat{f}}_k.{\widehat{f}}_k^{\text{'}}.{\widehat{f}}_k^{\text{'}\text{'}}\right)$

where c is the speed of light, f _{s is} the sampling frequency and B is the radar bandwidth.

LIST OF REFERENCE NUMBERS

1

vehicle

2

object

3

radar sensor

h _s

sensor height

h _t

height

R

distance

v

speed

v _r

radial velocity

v _s

sensor speed

v _x

velocity component

v _y

velocity component

w

yaw rate

x

axis

x '

axis

y

axis

y '

axis

β

Sensor mounting bracket

ε

elevation angle

θ

azimuth

θ _s

Sensor azimuth angle

QUOTES INCLUDE IN THE DESCRIPTION

This list of the documents listed by the applicant has been generated automatically and is included solely for the better information of the reader. The 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

• DE 102015009382 A1 [0002]

Cited non-patent literature

• "J. Li and P. Stoica: Efficient mixed-spectrum estimation with applications to target feature extraction; Signal Processing, IEEE Transactions on, vol. 44, no. 2, pp. 281-295, 1996 "[0028]

Claims (1)

Method for the radar-based measurement and / or classification of objects (2) in a vehicle environment, wherein - the vehicle surroundings are detected by means of at least one radar sensor (3) arranged on a vehicle (1) and - in a determination and / or classification of a height (h _t ) of an object (2) are generated on the basis of an evaluation of a shift of a Doppler frequency between a Radarsensor (3) emitted and one of the object (2) radar signal Doppler information, characterized in that - the determination and / or classification of the height (h _t ) of the object (2) is performed by means of at least one trained neural network, wherein - a training of the neural network based on simulation data is generated, which are generated from the Doppler information.

Images