Determining a warning or intervention time to avoid vehicle collisions

Abstract

Method for providing a time-difference probability density (p(t) Δ )), namely the probability density for time differences (t Δ ) to avoid a collision between an ego vehicle and another road user, the procedure including: Providing basic data, namely the distance of the ego vehicle (x 0,ego ) and the other road user (x 0,obj ) from the possible collision area, the speed of the ego vehicle (v 0,ego ) and the other road user (v 0,obj ) the acceleration of the ego vehicle (a 0,ego ) and the other road user (a 0,obj ); where the time difference (t Δ ) depends on the basic data; Determining a domain ( D t Δ ) for one or more basic data points, in particular for the distance of the ego vehicle (x ego), namely a range of values ​​for the basic data(s) in which the time difference (t) Δ ) a predetermined criterion is met, Determining a mean value (µ) depending on I time differences (t) Δ ), each representing a basic data point or a combination of basic data points, in particular a distance (x i,ego ) of the Ego vehicle, are calculated, whereby the I basic data or combinations of basic data are taken into account the domain of definition. ( D t Δ ) be determined; Determining a standard deviation (σ) depending on the time differences (t) Δ ); Determining the time-difference probability density (p(t) Δ )) depending on the determined mean (µ) and the standard deviation (σ).

Description

[0001] The invention relates to a method for providing a time difference probability density, namely the probability density for time differences by which an ego-vehicle and another road user avoid a collision. Based on the time difference probability density, the probability that the time difference lies below a threshold can be calculated. Based on this, a decision can be made as to whether a warning or a driving intervention should occur.

[0002] Previous driver assistance systems were mostly designed to prevent the severity of an unavoidable collision, for example, by applying the brakes late. Driver assistance systems are also under development that intervene as soon as a collision is detected. This intervention can take the form of steering, braking, and / or acceleration maneuvers. In addition to actual collisions with other road users, near misses are also considered critical traffic situations and should be avoided. Critical traffic situations are typically defined as situations in which the trajectories of road users intersect, but the road users do not actually collide. These road users pass through the collision zone—the area where the trajectories intersect—in quick succession.The time difference by which road users miss each other is called the time difference t. Δ This is referred to as a time gap (sometimes also called a time gap). A time gap of 0 s usually symbolizes a collision. The smaller the time gap t, the more Δ The longer the gap, the more critically the traffic situation is perceived by road users. A warning or intervention by a driver assistance system can therefore be considered useful even when the time gap falls below a certain threshold.

[0003] In practice, collision avoidance systems rely on inaccurate measurements of the traffic situation. These inaccuracies affect, for example, the absolute position determination of the ego vehicle, which is carried out using satellite navigation, and the relative position determination to other road users, which is carried out using vehicle sensors such as radar and / or lidar. Furthermore, inaccuracies can occur in the measurements of other conditions, such as speed and acceleration.

[0004] To account for measurement inaccuracies, it can be assumed, for example, that measured values ​​are subject to a normally distributed error. This leads to probabilistic statements regarding the locations and collision probabilities of vehicles when assessing a traffic situation. Calculating these probabilistic statements can be very computationally intensive and may not be achievable in real time with the processing units available in current vehicles and those expected in the future. However, real-time capability is a mandatory requirement for the use of driver assistance systems. For this reason, approximate solutions for calculating probabilistic statements have been proposed that are less computationally intensive than calculating the original function.To calculate probabilities for measurements with normally distributed errors, the method of "unscented transformations" was proposed, as described, for example, in the dissertation by U. Lerner "Hybrid Bayesian Networks for Reasoning about Complex Systems", Stanford University 2002, or in Julier, SJ; Uhlmann, JK: Unscented filtering and nonlinear estimation. In: Proceedings of the IEEE 92 (2004), pp. 401-422.

[0005] When calculating time differences, as explained above, the problem often arises that the calculation leads to singularities. This is understandable, for example, in the case where a vehicle decelerates (brakes) at a known rate and, depending on the assumed measurement error for the vehicle's position, sometimes comes to a stop before the collision zone and sometimes passes through the collision zone. In the first case, there is an infinite or undefined time difference; in the second case, a continuous time difference according to a relationship that can be described by a formula. Based on such pronounced time differences, the method of "unscented transformations," for example, cannot be applied.

[0006] DE 10 2013 005 362 A1 discloses a method for analyzing a traffic situation between at least one vehicle and at least one other road user at road intersections or junctions. Hazard zones are dynamically defined based on the intersections of predicted movement trajectories, whereby the times until arrival at and departure from the respective hazard zone are determined, evaluated, and taken into account for the vehicle and the relevant road users.

[0007] DE 10 2011 017 323 A1 also discloses a method for determining, within the vehicle, the probability of a collision between a vehicle and an object.

[0008] It is therefore the object of the present invention to provide a method for calculating probability statements about time differences that can be operated with computational effort suitable for real-time application in vehicles.

[0009] The object of the invention is achieved by a method, a control unit, and a vehicle according to the independent claims. Advantageous embodiments are defined in the dependent claims.

[0010] A first aspect of the invention relates to a method for providing a time-difference probability density (p(t) Δ )), namely the probability density for time differences (t Δ ), in order for an ego-vehicle and another road user to avoid a collision, the method comprising: providing basic data, namely the distance of the ego-vehicle (x 0,ego ) and the other road user (x 0,obj) from the possible collision area, the speed of the ego vehicle (v 0,ego ) and the other road user (v 0,obj ) the acceleration of the ego vehicle (a 0,ego ) and the other road user (a 0,obj ) (Additionally, the basic data can also include the standard deviation of the errors in the distance measurement of the ego and other vehicles (σ) o,x,ego ; σ 0,x,obj ) include); where the time difference (t Δ ) depends on the basic data; Determining a domain (DtΔ) for the removal of the Ego vehicle (x ego ), namely a range of values ​​in which the time difference (t Δ ) a predetermined criterion is met, determining a mean value (µ) depending on I time differences (t) Δ ), each for a distance (x i,ego ) of the Ego vehicle are calculated, where the I distances (x i,ego ) taking into account the domain of definition (DtΔ) to be determined and, in particular, additionally taking into account the standard deviation σ 0,x , which is determined based on the measurement errors of the distance of the ego vehicle; determining a standard deviation (σ) depending on the time differences (t) Δ );Determining the time difference probability density (p(t) Δ )) depending on the determined mean (µ) and standard deviation (σ). The distances I (x) are used. i,ego ) in particular using Gaussian quadrature or the method of exact monomials. The domains of definition only need to be determined for those states or basic data that are subject to uncertainties or are assumed to be.

[0011] In a training course that assumes uncertainties not only in the distance of the ego vehicle but also in the distance of the other vehicle (other road user) (in some implementations the standard deviation σ can also be taken into account). 0,x,obj (The uncertainty, which is determined based on the measurement errors of the distance of the other vehicle, is taken into account), the range of definition is also determined for distances of the other road user, and the time differences (t) are used to determine the mean value. Δ ) each for one of I combinations of distances {x i,ego , x i,obj} of the ego vehicle and the other road user is calculated, where the I combinations {x i,ego ,x i,obj} taking into account the domain (DtΔ) be determined.

[0012] This proposes determining the range of the time difference that fulfills a predetermined criterion, for example, exists and is continuous depending on the position of the ego-vehicle x. 0,ego and the other road user x 0,obj This requires determining the range of values ​​for the position of the ego vehicle and the other road user for which the criterion is met. These ranges of values ​​then constitute the domain. DtΔ. For this domain of definition, an approximation of the probability density of the time difference is then performed by defining support points {x i,ego , x i,objThe probability density function can be determined based on the data points. The mean and standard deviation of the probability density function are then calculated based on these data points. The probability density function can also be determined in two parts, i.e., using two probability density functions. For example, one probability density function can be determined for the case where the time difference is positive, and another for the case where the time difference is negative. The appropriate combination (combination) of both partial probability density functions then yields the overall probability density function for both positive and negative time differences.

[0013] This proposal therefore suggests using an approximation by approximating the mean and standard deviation based solely on the I combinations of ego-vehicle and road user positions. Simultaneously, the approximation is only performed for combinations of ego-vehicle and road user positions for which the time difference fulfills a predetermined criterion. This predetermined criterion can be the continuity of the time difference function and the absence of singularities. In this way, the feasibility of the approximation is ensured, and thus the real-time capability of the calculation.

[0014] The exact monomial method makes it possible to determine an approximation for the mean and standard deviation of the transformed distribution (or the time gap). Conversely, the structure of the calculation—namely, first the approximate determination of the mean and standard deviation before determining the approximate probability density—corresponds to the application of Gaussian quadrature or the exact monomial method. These applications are based on the evaluation of functions at support points defined by the I combinations of the positions of the ego vehicle and the road users.

[0015] Furthermore, it can be stipulated that if the time gap / difference is small, it must also be determined that the remaining time to avoid a collision through any driver intervention (time to react - TTR) must also be small. For individual driver interventions, various final possible execution times can be determined: time to break (TTB), time to steer (TTS), time to kickdown (TTK), or more generally, time to thru (TTX). By considering the TTR, a time-based criterion is used to assess the criticality of a situation and to execute an intervention / warning. A small time gap is therefore a necessary but not yet sufficient criterion for driver intervention. The calculation of the TTX parameters can be performed as described in the following documents: Hillenbrand, J.: Driver Assistance for Collision Avoidance, Faculty of Electrical Engineering and Information Technology, University of Karlsruhe, Dissertation, 2007; and Tamke, A.; Dang, T. ; Breuel, G.: A Flexible Method for Criticality Assessment in Driver Assistance. In: IEEE Intelligent Vehicles Symposium (IV), 2011.

[0016] In an advantageous implementation, determining the mean (µ) includes: determining I weighting factors (w) i ), in particular by means of Gauss-Hermite, Gauss-Legendre or Gauss-Laguerre quadrature, or the method of exact monomials (the choice of one of the Gaussian methods is made depending on whether the domain is not bounded at all, one-sided or two-sided); Where the weighting factors (w) are used to determine the mean i ) are taken into account.

[0017] In another advantageous implementation, determining the mean (µ) includes: determining I weighting factors {w i} and I combinations of auxiliary support points {xi,egoGL,xI,objGL}, in particular by means of Gauss-Hermite, Gauss-Legendre or Gauss-Laguerre quadrature or the method of exact monomials (the choice of one of the Gaussian methods is made depending on whether the domain of definition is not bounded at all, one-sided or two-sided); where the I combinations of the distances ({x i,ego ,x i,obj}) each depending on one of the I combinations of the support points ({xi,egoGL,xI,objGL} and the domain (DtΔ) be determined; where the mean is determined by a summation in which the weighting factors ({w i ,}) are used as factors of the summands. In the implementation of the approximation, weighting factors {w} are used. iThe auxiliary support points are calculated for each of the support points and used in the summation of the approximation. Ideally, the auxiliary support points themselves are determined using tables such as those described in Abramowitz, Milton; Stegun, Irene A.: “Handbook of mathematical functions: with formulas, graphs, and mathematical tables”, Courier Dover Publications, 2012.

[0018] The particular advantage of Gaussian quadrature or the exact monomial method over other approximations lies in the fact that the support points are advantageously distributed over a large portion of the domain and their calculation can be performed efficiently using pre-calculated (stored) intermediate values. This type of evaluation thus enables a particularly good approximation of the probability density.

[0019] Just like the mean, the standard deviation (σ) can also be determined by a summation where the weighting factors (w) i ) be used.

[0020] In advantageous advanced training, the procedure further includes: Determining the cumulative probability density (p(|t)). Δ | < t Δ,min )) which gives the probability that the magnitude of the time difference is less than a minimum value (t Δ,min ) where the cumulative probability density is derived from the probability density of the time difference, in particular by summation or integration. The cumulative probability density indicates the probability that the ego vehicle and the other road user will miss each other or collide by less than the minimum value. Computationally efficient and therefore real-time capable implementations exist for calculating the cumulative probability densities.

[0021] Depending on this probability, warnings or control interventions can be carried out when the cumulative probability (p(|t) Δ | < t Δ,min )) is greater than a warning threshold.

[0022] As already indicated above, the cumulative probability can be calculated in two parts; where a mean value (µ) is calculated for each part. Δ,p , µ Δ,n ) and a standard deviation (σ Δ,p , σ Δ,n ) is determined. Typically, the basic data also includes the length of the ego vehicle (l). ego ) (herein also referred to as l ego,fzg designated), the length of the other road user (l obj ) (herein also referred to as l obj,fzg designated) as well as the width of a passage through the collision area (l kb ), which exist for both the Ego vehicle and the foreign vehicle, l obj,kb and l ego,kbThis information is needed to determine the collision area and time differences more precisely.

[0023] Two further aspects include a control unit, comprising electronic computing means, wherein the control unit is configured to perform one of the previously described procedures; and a motor vehicle with such a control unit. BRIEF DESCRIPTION OF THE DRAWINGS Fig. Figure 1 schematically shows a possible collision situation between two vehicles according to an exemplary embodiment. Fig. Figure 2 shows an example of how to calculate a probability density for the time gap. Fig. 3 (a) and (b) show another example of calculating the probability density. DETAILED DESCRIPTION OF THE EXECUTION EXAMPLE

[0024] Fig. Figure 1 schematically shows a possible collision situation between two vehicles according to an exemplary embodiment. Fig.Figure 1 shows the ego vehicle 1 and the other road user 2. A route path is predicted for both vehicles based on their previous movements. The distance from the ego vehicle's current position on the route path is denoted by x, and here also by x. ego The same applies to the other road user 2. With x0 or x 0,ego The distance to entering collision zone 3 is described. With l kb The distance that must be traveled in addition to x0 to leave the path of the other road user is called x0. At the current point, the ego vehicle has a speed v0 and an acceleration a0. The ego vehicle's entry time into the collision zone can be expressed as x0. tein=1a0(−υ0+υ02+2a0x0)︸=:ft(x0,υ0,a0) described herein also as t ego,ein designated. Likewise, the exit time of the ego vehicle t ego,aus, the entry time of the other road user, t obj,ein , and the exit time of the other road user t obj,aus The calculation of the entry and exit times reveals that each of the four times considered can have complex values. This corresponds, intuitively, to the case where a vehicle comes to a standstill before or within the conflict zone. The time gap is only defined if both vehicles reach the conflict zone, i.e., if both entry times are real. The exit times, however, can have complex values: If a vehicle comes to a standstill within the conflict zone, the time gap is real and non-zero if and only if the other vehicle has left the conflict zone before the vehicle in question enters. Otherwise, the time difference becomes zero, as it does in the case where both exit times have complex values.

[0025] The time difference is then calculated as follows: tΔ={tego,ein−tobj,aus,wenn tobj,aus∈ℝ, tego,ein>tobj,austego,aus−tobj,ein,wenn tobj,aus∈ℝ, tego,ein>tobj,aus0,sonst,

[0026] The domain is defined as follows: DtΔ={{tego,in,tobj,in}∈ℝ2,{tego,out,tobj,out}∈ℂ2}

[0027] As explained in the introduction, issuing a warning to the driver and / or initiating a driving intervention can already be useful if the time difference is less than a threshold value t. Δ,min This is because otherwise the traffic situation would be considered critical. For this reason, the probability p(|t) Δ | < t Δ,min , t Δ ∈ ℝ) calculates that the magnitude of the time difference is defined and less than the threshold t Δ,min is. Provided that this probability p(|t Δ | < t Δ,min t ΔIf the probability of the vehicle exceeding a threshold value (∈ ℝ) is again triggered, a warning is issued and / or a driving intervention is carried out.

[0028] Applying Bayes' theorem, the following holds true: p(|tΔ|≤tΔ,min, tΔ∈ℝ)=(|tΔ|≤tΔ,min|tΔ∈ℝ)⋅p(tΔ∈ℝ)

[0029] The two factors of (Eq. 1) can thus be determined separately.

[0030] The second part of (Eq. 1) can be transformed as follows: p(tΔ∈ℝ)=p(tego,a∈ℝ,tobj,a∈ℝ) =p((vego2+2aego,0xego,0≥0),(vobj,02+2aobj,0xobj,0≥0))

[0031] Assuming that the measurement noise of x ego,0 and x obj,0 being statistically independent leads to =p(vego,02+2 aego,0 xego,0≥0)∗p(vobj,02+2 aobj,0 xobj,0≥0).

[0032] The two factors can therefore be determined separately and analogously for the ego vehicle and the other vehicle. For the ego vehicle example, this leads to p(vego,02+2 aego,0 xego,0≥0)={xego,0≤−vego,022aego,0,aego,0<0xego,0≥−vego,022aego,0,aego,0≥0 The parameters of the case distinction can be determined by evaluating the cumulative probability distribution. =Φ(−vego,022aego,0|xego,0, σego,02)

[0033] Here, Φ(x|µ, σ) denotes 2 ) the cumulative distribution function of the normal distribution, which is obtained from the integral over the normally distributed probability density with expected value µ and variance σ 2 results Φ(x|μ,σ2)=∫−∞xN(z|μ,σ2)dz

[0034] Furthermore, the abbreviated notation is used to determine the first part of (Eq. 1): p(|tΔ| <tΔ,min|tΔ∈ℝ)=p(|tΔ|<tΔ,min)

[0035] The resolution of the absolute value bars by dividing the distribution of the time gap into a positive and a negative section remains valid. p(|tΔ|≤tΔ,min)=1−p(|tΔ|>tΔ,min)=1−p(tΔ,n<−tΔ,min)−p(tΔ,p>tΔ,min)

[0036] Using the above definition of time difference, it follows that the time gap is only positive if (t Δ,p ) or is negative (t Δ,n ), if the first or second case of the case distinction occurs. Therefore, the two quantities t Δ,p and t Δ,n as two separate distributions. For example, this results in the following: tΔ,p=tego,in−tobj,out=ft(xego,0,vego,0,aego,0)−ft(xobj,out,vobj,0,aobj,0).

[0037] where x obj,aus = x obj,0 + l obj,fzg +l obj,kb

[0038] The cumulative probability densities are determined as follows: p(tΔ,p>tΔ,min)=1−Φ(tΔ,min|μΔ,p,σΔ,p2) p(tΔ,n<−tΔ,min)=Φ(−tΔ,min|μΔ,n,σΔ,n2).

[0039] Here, Φ(x|µ, σ) denotes 2 ) the cumulative distribution function of the normal distribution, which is obtained from the integral over the normally distributed probability density with expected value µ and variance σ 2 results Φ(x|μ,σ2)=∫−∞xN(z|μ,σ2)dz

[0040] The cumulative probabilities Φ(tΔ,min|μΔ,p,σΔ,p2) and Φ(−tΔ,min|μΔ,n,σΔ,n2) are now advantageously approximated using the method of exact monomials according to the invention. In the further course of this, the x i and w iListed for accuracy level 3. Higher-order methods are described in U. Lerner, "Hybrid Bayesian Networks for Reasoning about Complex Systems," Stanford University, 2002, or in Julier, S.J.; Uhlmann, J.K.: Unscented filtering and nonlinear estimation. In: Proceedings of the IEEE 92 (2004), pp. 401–422.

[0041] The following applies (assuming that the position measurement errors of the ego vehicle and the other road user are independent): μΔ,p=E(tΔ,p)=w0*t0,Δ,p+∑i=1Iw1*ti,Δ,p σ2Δ,p=E2(tΔ,p)−μ2Δ,p where E2(tΔ,p)=w0*t20,Δ,p+∑i=1Iw1*t2i,Δ,p (Accordingly, the values ​​for t can be determined.) Δ,n calculate.)

[0042] Here, I denotes the number of support points - 1, where I = 4.

[0043] Furthermore, the following applies: w0=1−du2with d=2(dimension of μΔ,p) w1=12u2

[0044] Where umax=min(uego,uobj) uego=(xego,lim−x0,ego) / σego uobj=(xobj,lim−x0,obj) / σobj

[0045] where u is in the range [0 ... u max ] is chosen to ensure that all function evaluations are defined. This requires u max be sufficiently large. Often u=3 chosen, compare Lerner, U.: Hybrid Bayesian Networks for Reasoning about Complex Systems, Stanford University, Dissertation, 2002. For the choice of u=sqrt(3), the procedure corresponds to that of the Unscented Transformation.

[0046] If u max If the data is very small, the control points usually need to be scaled up significantly. In this case, the exact monomial method can be modified and supplemented by Gaussian quadrature.

[0047] Generally, for each page of t Δ (t Δ,n and t Δ,p , see below) a u max certainly.

[0048] This results in x ego,limand x obj,lim according to the following table: f Condition for f ∈ ℝ t ein x0∈{[−12v02a0,∞)if a0>0(−∞,−12v02a0]if a0<0 t aus x0∈{[−12v02a0−lfzg−lkb,∞),if a0>0(−∞,−12v02a0−lfzg−lkb],if a0<0 σ ego and σ obj are the standard deviations of the measurement inaccuracies of the distance determination of the ego vehicle and the other road user.

[0049] For t i,Δ,p applies: ti,Δ,p=ti,Δ,p(xi) With xi=[σego00σobj]∗ui+[x0,egox0,obj] Where u0=

[00] and ui is selected from {[−u 0],[u0],[0−u],[0u]}, with i=1,2,3,4.

[0050] Fig. Figure 2 shows an example of how to calculate a probability density for the time gap. The total probability density p(t) Δ ) results from the probability densities for t Δ,n and t Δ,p . In Fig.Figure 2 shows the overall probability density function as a step function. In this example, there is a certain probability that the ego vehicle leaves the conflict area before the other vehicle reaches it (time gap < 0 s). At the same time, it cannot be ruled out that the vehicles pass through the conflict area in reverse order (time gap > 0 s).

[0051] Fig. Figures 3 (a) and (b) show a second example of calculating the probability density, here only for the proportion t. Δ,p The domain lies below the dashed line. Furthermore, a comparison of two approximation methods for the probability density function of t is presented. Δ,p shown. On the one hand, using the method of exact monomials with accuracy degree 5 and 9 support points, abbreviated by exMo5 (circles in Fig. 3 (b)). How Fig.As shown in Figure 3(B), support points symmetrically distributed around the expected value must be strongly compressed to avoid function evaluations outside the domain. This means the function is only evaluated in a very limited area close to the expected value. 0,ego , x 0,obj Therefore, the actual distribution is only evaluated with unsatisfactory accuracy (see Fig. 3 (a)). An increase in approximation accuracy is made possible by applying the method according to the invention. Due to the one-sided limited integration interval, Gauss-Laguere integration is used. In Fig. 3 (a) and (b) is the result of the combined Gauss-Laguerre integration for the state x obj and Gauss-Hermite integration of order 3 for the state x ego This method is abbreviated as G.Lag / ExMo3 and uses the same number of support points as exMo5. How Fig.Figure 3(b) shows that the support points are along the state x obj no longer arranged symmetrically around the expected value. Fig. 3 (a) shows that this method approximates the positive portion of the time gap distribution much more accurately. At the same time, this method enables real-time execution on current and anticipated future vehicle computing units.

Claims

[1] Method for providing a time-difference probability density (p(t Δ )), namely the probability density for time differences (t Δ ) to avoid a collision between an ego vehicle and another road user, the procedure including: Providing basic data, namely the distance of the ego vehicle (x 0,ego ) and the other road user (x 0,obj ) from the possible collision area, the speed of the ego vehicle (v 0,ego ) and the other road user (v 0,obj ) the acceleration of the ego vehicle (a 0,ego ) and the other road user (a 0,obj ); where the time difference (t Δ ) depends on the basic data; Determining a domain (DtΔ) for one or more basic data points, in particular for the distance of the ego vehicle (x ego), namely a range of values ​​for the basic data(s) in which the time difference (t) Δ ) a predetermined criterion is met, Determining a mean value (µ) depending on I time differences (t) Δ ), each representing a basic data point or a combination of basic data points, in particular a distance (x i,ego ) of the Ego vehicle, are calculated, whereby the I basic data or combinations of basic data are taken into account the domain of definition. (DtΔ) be determined; Determining a standard deviation (σ) depending on the time differences (t) Δ ); Determining the time-difference probability density (p(t) Δ )) depending on the determined mean (µ) and the standard deviation (σ). [2] Method according to claim 1, where the domain (DtΔ) for the removal of the Ego vehicle (x ego) and the distance of the other road user (x obj ) is determined, where the I time differences (t Δ ) each for one of I combinations of distances {x i,ego , x i,obj} of the ego vehicle and the other road user are calculated, where the I combinations (x i,ego , x i,obj} taking into account the domain (DtΔ) be determined. [3] Method according to claim 1, wherein the determination of the mean value (µ) comprises: Determining I weighting factors (w i ) using Gauss-Legendre, Gauss-Laguerre or Gauss-Hermite quadrature or the method of exact monomials; Where the weighting factors (w) are used to determine the mean i ) are taken into account. [4] Method according to claim 2, wherein the determination of the mean value (µ) comprises: Determining I weighting factors {w i} using Gauss-Legendre, Gauss-Laguerre or Gauss-Hermite quadrature or the method of exact monomials; Where the I combinations of distances ({x i,ego , x i,obj}) each depending on the domain of definition (DtΔ) be determined; Where the weighting factors ({w) are used to determine the mean i}) be used. [5] Method according to claim 4, wherein the determination of the standard deviation (σ) is carried out by a summation in which the weighting factors ({w i}) are used as factors of the summands. [6] Method according to any one of the preceding claims, further comprising: Determining the cumulative probability density (p(|t) Δ | < t Δ,min )) which gives the probability that the magnitude of the time difference is less than a minimum value (t Δ,min ) is. [7] Method according to claim 6, further comprising: Issuing a warning and, in particular, executing a control intervention in the driving of the ego vehicle, if the cumulative probability (p(|t Δ | < t Δ,min )) is greater than a warning threshold. [8] Method according to claim 7, wherein the cumulative probability is calculated in two parts; wherein for each part a mean value (µ) Δ,p , µ Δ,n ) and a standard deviation (σ Δ,p , σ Δ,n ) is determined. [9] Method according to any one of the preceding claims, wherein the scope of definition (DtΔ) is calculated depending on the provided basic data. [10] Method according to one of the preceding claims, wherein the basic data provided also includes the length of the ego vehicle (l ego ), the length of the other road user (l obj ) as well as an initial curvature correction for the trajectories (l kb) include, in particular, the length of the collision area. [11] Control unit comprising electronic computing means, wherein the control unit is configured to execute a method according to any one of claims 1 to 8. [12] Motor vehicle with control unit according to claim 11.

Images