WO2020074195A1

WO2020074195A1 - Method for assessing a traffic situation

Info

Publication number: WO2020074195A1
Application number: PCT/EP2019/074070
Authority: WO
Inventors: Rene Wirtz
Original assignee: Daimler Ag
Priority date: 2018-10-10
Filing date: 2019-09-10
Publication date: 2020-04-16
Also published as: DE102018008024A1

Abstract

The invention relates to a method for assessing a traffic situation for a vehicle which is operated in an at least partially autonomous manner. According to the invention, an artificial neural network (2) is trained with data (D) which are captured by means of a vehicle's own environmental sensor system (3) during a non-critical traffic situation (V) which did not result in an accident involving the vehicle and/or in manual intervention by a driver during the at least partially autonomous operation of the vehicle. A future traffic situation (V_praed) is predicted by means of the trained artificial neural network (2), wherein, in the event of a significant deviation of an actually occurring traffic situation (V) from the predicted traffic situation (V_praed), a critical traffic situation (V_krit) is inferred, and, if there is a critical traffic situation (V_krit), an intensity of an operation of monitoring at least one road user, the behaviour of which significantly differs from a predicted behaviour, which is carried out using the vehicle's own environmental sensor system (3), is increased.

Description

Procedure for evaluating a traffic situation

The invention relates to a method for evaluating a traffic situation according to the preamble of claim 1.

DE 10 2016 007 899 A1 discloses a method for operating a device for traffic situation analysis, which processes input data relating to a partially autonomously operable motor vehicle. The input data include geographic map data describing the surroundings of the motor vehicle and the positions and directions of movement of other road users in the surroundings of the vehicle

Ambient data describing the motor vehicle. The process has the following steps:

Determine a predicted range of motion of the motor vehicle

descriptive movement space information from the map data;

Determining movement path information describing a predicted movement path of a road user by applying one

Motion prediction model on the surrounding data;

Determine the relevance for a traffic situation analysis of a

Partial movement space descriptive evaluation information depending on the movement space information and the movement path information; and

Carrying out the traffic situation analysis for the partial movement area in

Dependency of the evaluation information.

The invention is based on the object of specifying a method for evaluating a traffic situation which is improved compared to the prior art.

The object is achieved by a method which the in

Features specified claim 1.

Advantageous embodiments of the invention are the subject of the dependent claims. According to the invention, an artificial neural network is used to evaluate a traffic situation for an at least partially autonomous vehicle during an uncritical traffic situation which does not result in an accident of the driver

Vehicle and / or has not led to a manual intervention by a driver during the at least partially autonomous operation of the vehicle by means of a

vehicle-acquired environment sensors trained data. The training enables the neural network to predict a future traffic situation.

Using the trained artificial neural network, the future

Traffic situation predicted, in the event of a significant deviation of a real traffic situation from the predicted traffic situation, a critical traffic situation is inferred and, in the presence of a critical traffic situation, an intensity of at least one traffic participant whose behavior differs significantly from a predicted behavior by means of the vehicle's own environmental sensor system, is increased.

Advantageously, the training of the neural network is first carried out on the basis of generic data which have previously been recorded by the environmental sensor system in non-critical traffic situations, and only then on the basis of the data which are recorded by the environmental sensor system during operation of the vehicle.

The method enables an improvement in the evaluation of a traffic situation due to the increase in the intensity of the monitoring. This results in the possibility of recognizing the risk of an accident at an early stage and thus increasing traffic safety.

Exemplary embodiments of the invention are explained in more detail below with reference to a drawing.

It shows:

Fig. 1 shows schematically a block diagram of a device for evaluating a

Traffic situation. Figurh Figure 1 is a block diagram of a possible embodiment of a device 1 for evaluating a traffic situation for an at least partially autonomous vehicle is shown.

The device 1 comprises an artificial neural network 2, an in-vehicle environmental sensor system 3 with a number of sensors 3.1 to 3.n, for example a camera, a radar sensor, a lidar sensor, an ultrasonic sensor and / or other sensors, an evaluation unit 4 and a control unit 5 to control the environmental sensors 3.

For reliable autonomous or semi-autonomous operation of the vehicle, it is beneficial to have knowledge of future traffic situations V. For this it is necessary to predict a future traffic situation V.

This is done by means of a trained artificial neural network 2, which is used during an uncritical traffic situation V, which does not lead to an accident of the

Vehicle and / or has not led to a manual intervention by a driver during the at least partially autonomous operation of the vehicle, by means of which the data D recorded in the vehicle's own environmental sensor system 3 is trained. The data D of the sensors 3.1 to 3.n, which are required as input for the execution of the at least partially autonomous ferry operation, are evaluated by the artificial neural network 2 in such a way that this predicts a movement of road users detected in the surroundings of the vehicle into the vehicle Can carry out future. The result of this prediction is a predicted future traffic situation V _pra ed.

The predicted future traffic situation V _pra ed is compared together with a real traffic situation V likewise determined on the basis of the data D acquired by the environmental sensor system 3, information about the predicted future traffic situation V _prae d and the real traffic situation V being supplied to an evaluation unit 4 for this purpose. In one possible embodiment of the device 1, the evaluation unit 4 can be part of the artificial neural network 2. In this

The comparison compares a real behavior of other road users continuously at a time t + Ät, where Ät corresponds to a prediction period or prediction interval, with a behavior predicted at time t. Comparison of a significant deviation from the real occurring

verKenrssituation V determined by the predicted traffic situation V _pra ed, _crit is closed to a critical traffic situation V. That is, if there is a significant deviation between the real behavior and the predicted behavior, which is based entirely on information from the traffic situation V _pra ed predicted on the basis of uncritical traffic situations V, an abnormal behavior of the observed road user is assumed and the critical behavior

Traffic _situation V _crit closed.

In this case, information about the critical traffic _situation V _{crit is sent} to the control unit 5, which controls the environmental sensor system 3 in such a way that an intensity of a monitoring of the at least one traffic participant, which is performed by means of the environmental sensor system 3 and whose behavior differs significantly from the predicted behavior, is increased. This means that the environmental sensor system 3 can use larger computing resources for observing the road user, in order to detect an accident risk as quickly as possible and in order to control

If necessary, activate vehicle functions at an early stage to take safety measures to reduce the risk of accidents and / or the consequences of accidents. In this case, for example, at least one high-resolution sensor 3.1 to 3.n movably arranged on the vehicle with a small opening angle, for example a camera, a radar sensor or

Lidar sensor, the vehicle's own environmental sensor system 3 aimed at the

Road users are directed or a grid of a local and temporal frame is reduced in a calculation of a movement of the road user, d. H. a finer local and time frame, also known as a grid, in which

Calculation of the movement of the road user is used.

Furthermore, an optimal prediction period Ät is determined, which is small enough to produce a prediction that is as exact as possible, but is not too short for possible ones

Detect deviations in real behavior from prediction. The size of the prediction period Ät becomes variable depending on the vehicle's own

Environment sensors 3 set ambient conditions. This means that the prediction period Ät is designed to be variable, so that it is adapted to the external conditions, for example a speed of the vehicle and / or a range of vision of the sensors 3.1 to 3.n.

Claims

1. Method for evaluating a traffic situation for an at least partially autonomously operated vehicle,

characterized in that

- An artificial neural network (2) with during an uncritical

Traffic situation (V), which did not lead to an accident of the vehicle and / or to a manual intervention by a driver during the at least partially autonomous operation of the vehicle, by means of an in-vehicle

Environment sensors (3) recorded data (D) is trained,

a future traffic situation (V _pra ed) is predicted by means of the trained artificial neural network (2),

- In the event of a significant deviation of a real traffic situation (V) from the predicted traffic situation (V _pra ed) to a critical one

Traffic situation (Vkrit) is closed and

- In the presence of a critical traffic situation (V ^ _t ), an intensity of a monitoring carried out by means of the vehicle's own environmental sensor system (3) is increased at least one traffic participant whose behavior differs significantly from a predicted behavior.

2. The method according to claim 1,

characterized in that

the intensity of the monitoring is increased in that at least one high-resolution sensor (3.1 to 3.n) of the vehicle's environmental sensor system (3), which is movably arranged on the vehicle, is directed at the road user.

3. The method according to claim 1 or 2,

characterized in that

the intensity of the monitoring is increased by using a grid dimension of a local and time frame in a calculation of a movement of the

Road user is reduced.

4. The method according to any one of the preceding claims, characterized in that

a size of a prediction period is variably set depending on the environmental conditions detected by the vehicle's own environmental sensor system (3).