DE102018002029A1 - Method for determining a prediction for a pose of a vehicle for driver assistance systems - Google Patents

Abstract

The invention relates to a method for determining a prediction for a pose of a vehicle for driver assistance systems, in which past measurement data (5) of a measuring arrangement (3) of the vehicle are recorded and processed. In a method in which a simple but very accurate prediction of the pose of the vehicle is predicted, the past measurement data (5) are learned by means of a neural network (1), the neural network (1) being derived from the learned measurement data (5). determines a future pose of the vehicle.

Description

The invention relates to a method for determining a prediction for a pose of a vehicle for driver assistance systems, in which past data of a measurement arrangement of the vehicle are recorded and processed.

For driver assistance systems in vehicles, in particular for head-up displays and head-mounted devices, it is of particular importance that a precise position and orientation of the vehicle is known. The accuracy must be chosen so that a finished rendered image fits exactly in time with the display time.

From the DE 10 2016 211 805 A1 a fusion of position data by means of Posengraph is known in which absolute position data of the vehicle and odometry position data of the vehicle are determined. A posengraph is generated, with edges of the posengraph corresponding to the odometry position data and nodes of the posengraph to the absolute position data. Thereby, an estimated position is reached and the vehicle is controlled based on the estimated position. However, this proposed model only allows to estimate the current situation of the vehicle.

The object of the invention is to provide a method in which future Posen the vehicle can be accurately determined without much computational effort.

The invention results from the features of the independent claims. Advantageous developments and refinements are the subject of the dependent claims. Other features, applications and advantages of the invention will become apparent from the following description, as well as the explanation of embodiments of the invention, which are illustrated in the figures.

The object is achieved with a method in that the past measurement data are learned by means of a neural network, wherein the neural network determines from the learned measurement data a future pose of the vehicle. The advantage is that the neural network can recognize correlations by means of examples or training data, which are very difficult to modulate by hand. The neural network automatically learns the driving physics, natural frequencies and relationships of the vehicle, whereby it can adapt continuously to the variable characteristics of the vehicle through permanent learning. Learning with the neural network can be activated and deactivated at any time to save resources. The implementation of neural networks is very efficient and can even be more resource efficient than handcrafted models. In this training process irrelevant data is ignored, the neural network resource consumption is constant and depends mainly on the selected width and depth of the neural network.

Advantageously, after entering the past input data, the neural network trains at a current training time and determines future measurement data for the expected pose at a fixed future time. Due to this procedure, an expected result can be predicted with high reliability.

In one embodiment, a predetermined number of measurement data of a measurement arrangement of the vehicle, which were measured in a period prior to the current training time and entered into the neural network and processed by the neural network, the neural network is a prediction of the future measurement data for the pose of Vehicle at a fixed time after the training time spent. This makes it particularly easy to specify only a single value in the future.

In one variant, the neural network will train for input of the past input data at the current time and continue to carry characteristics of a series of input data in the future. Thus, not only individual values can be output, but a series of future values can be determined.

In a further development, a plurality of input data of the measuring arrangement of the vehicle are successively measured in time and supplied to the neural network, wherein the neural network assigns a measurement result determined for the future time for successive future times. Thus, a series started in the past can be seamlessly continued for the future.

In one embodiment, by buffering the input data, training of the neural network is timed to a live operation. This makes it easy to set a timeshift between such data.

In a further embodiment, a long-short-term memory (LSTM) is used as the neural network. In particular, such a long-short-term memory enables the processing of several consecutive input data of the past and the generation of data series of the future.

In a particularly simple embodiment, as a measuring arrangement of the vehicle Gyroscope and / or an accelerometer and / or a wheel tachometer and / or a GPS used. The measurement results of these measurement arrangements can be processed either individually or in combination by the neural network, allowing a more accurate prediction of a future pose of the vehicle.

Further advantages, features and details will become apparent from the following description in which - where appropriate, with reference to the drawings - at least one embodiment is described in detail. Described and / or illustrated features may form the subject of the invention itself or in any meaningful combination, optionally also independent of the claims, and in particular may also be the subject of one or more separate application / s. The same, similar and / or functionally identical parts are provided with the same reference numerals.

• 1 ein erstes Ausführungsbeispiel des erfindungsgemäßen Verfahrens,
• 2 ein weiteres Ausführungsbeispiel des erfindungsgemäßen Verfahrens.

Show it:

• 1 a first embodiment of the method according to the invention,
• 2 a further embodiment of the method according to the invention.

In 1 a first embodiment of the method according to the invention is shown in which the learning of a frequency response of gyroscope measurements is shown. The neural network designed as LSTM 1 takes from a gyroscope 3 measured input data 5 on. All these input data 5 were recorded in a time interval of (tx, t). The neural network 1 processes this past input data 5 of the gyroscope 3 at time t and gives a single measurement result 7 from which is to arrive at a given time (t, .DELTA.t).

A second embodiment of the method according to the invention is in 2 shown. Here are also input data of a gyroscope 3 the neural network 1 available for measurement and evaluation. However, the input data 5.1... 5.n have been measured at different points in time, so that a time t1 is a first measurement of the input data 5.1, a time t2 a second measurement of the input data is assigned 5.2 and so on until a time tn a measurement of last input data 5n. These measured data recorded at different times become the neural network 1 fed. By training this input data 5 recognizes the neural network 1 a pattern of input input data 5 and gives as output prediction data as a measurement result 7 from, wherein for the future time t + 1, a first predetermined measurement result 7.1 is determined, for the time t + 2, a second future measurement result 7.2 and for the time t + 3, a further future measurement 7.3 is determined. This is the series of input data 5 through the predictive measurement results 7 continued without interruption. This method can be executed as often as desired in order to obtain corresponding new forecast data series.

In addition to the gyroscope data, measurement data from other measurement arrangements of the vehicle can also be used as input data 5 for the neural network 1 be used. These include an accelerometer, a wheel tachometer, a GPS, but also a current result of the pose estimation, which by means of the neural network 1 can be made more concrete for the future. By means of the evaluation of these input data, it is possible to predict an actual future pose of the vehicle and to use this actually future pose in a head-up display or in head-mounted devices. A pose is understood to mean a combination of position and orientation of the vehicle, for example the current direction of movement of the vehicle.

Although the invention has been further illustrated and explained in detail by way of preferred embodiments, the invention is not limited by the disclosed examples, and other variations can be derived therefrom by those skilled in the art without departing from the scope of the invention. It is therefore clear that a multitude of possible variations exists. It is also to be understood that exemplified embodiments are really only examples that are not to be construed in any way as limiting the scope, applicability, or configuration of the invention. Rather, the foregoing description and description of the figures enable one skilled in the art to practice the exemplary embodiments, and those of skill in the knowledge of the disclosed inventive concept may make various changes, for example as to the function or arrangement of particular elements recited in an exemplary embodiment, without Protection area defined by the claims and their legal equivalents, such as further explanations in the description.

LIST OF REFERENCE NUMBERS

1

neural network

3

gyroscope

5

input data

7

measurement result

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 102016211805 A1 [0003]

Claims (8)

Method for determining a prediction for a pose of a vehicle for driver assistance systems, in which past measurement data (5) of a measurement arrangement (3) of the vehicle are recorded and processed, characterized in that the past measurement data (5) are generated by means of a neural network ( 1), wherein the neural network (1) determines from the learned measurement data (5) a future pose of the vehicle. Method according to Claim 1 , Characterized in that the neural network (1) to the input of the last input data (5) to a current training time (t) is trained and outputs future measurement data (7) for the expected pose at a fixed future point in time (t + .DELTA.t) , Method according to Claim 2 , characterized in that a predetermined number of input data (5) of a measuring arrangement (3) in a period (tx, t) before the current training time (t) of the neural network (1) measured and entered into the neural network (1) and from the neural network (1), the neural network (1) outputting a prediction of the future measurement data (7) for the pose of the vehicle at a fixed time (t + Δt) after the training time (t). Method according to Claim 1 characterized in that the neural network (1), after inputting the past input data (5), trains at the current training time (t) and continues characteristics of a series of input data (5) in the future. Method according to Claim 4 , Characterized in that consecutively a plurality of input data (5) of the measuring arrangement (3) of the vehicle is measured and the neural network (1) are fed, wherein the neural network (1) for consecutive future times (t + n) one for the future time (t + n) expected measurement result (7). Method according to at least one of the preceding claims, characterized in that by buffering the input data (5) the training of the neural network (1) takes place in a time-delayed manner to a live operation. Method according to at least one of the preceding claims, characterized in that a long-short-term memory (LSTM) is used as the neural network (1). Method according to at least one of the preceding claims, characterized in that a gyroscope (3) and / or an accelerometer and / or a wheel tachometer and / or a GPS are used as the measuring arrangement of the vehicle.

Images