DE102011108292A1 - Method for operating driver assistance device of vehicle, involves determining scenario-dependent sensor variances or sensor variances depending on driver assistance device in context of error propagation determination - Google Patents

Abstract

The method involves determining the scenario-dependent sensor variances or the sensor variances depending on the driver assistance device (6) in the context of error propagation determination in a functional validation of the sensors (2,3,4,5). The signal variances of the respective sensors are determined from the sensor variances.

Description

The invention relates to a method for operating a driver assistance device of a vehicle, in which an environment of the vehicle and / or vehicle-related data are detected by means of a plurality of sensors.

Modern driver assistance devices are characterized by high complexity due to high-bandwidth sensors, complementary sensor concepts, fusion of various sensors, environmental representation, situation analysis, detection of collision hazards through prediction, and a host of systemic responses. The increasing complexity of the driver assistance devices and the great potential for action of the driver assistance devices make high demands on a functional safety, ie. H. an avoidance of incorrect reactions of the driver assistance devices, at the same time largely maintaining a system efficiency, d. H. the draw in the case of use, the so-called "use case".

Current approaches to functional protection are often based on static methods, in which fixed threshold values are set for variables describing a scenario, such as an airspeed or a self-yaw rate, wherein a deactivation of the driver assistance device occurs when the threshold values are exceeded. The threshold values form scenario-specific limits of sensors or the entire driver assistance device. To functionally protect the sensors, in purely radar-based collision-reducing driver assistance devices, such as the so-called pre-safe brake, a brake assistant or active brake assistant, distance-dependent threshold values are checked for object confidence delivered by the radar. These threshold values are known as CMS levels (CMS: collision mitigation system).

From the DE 10 2004 035 578 A1 For example, a method for driving stabilization of a vehicle is known, wherein measured values that are suitable for determining a loading state of the vehicle are detected. On the basis of the measured values, a characteristic measure characterizing the loading state of the vehicle is formed. Furthermore, a control parameter set is selected and scaled in dependence on the characteristic measure, wherein the selected and scaled parameter set is transmitted to a driving stabilization control device of the vehicle. The driving stabilization control device stabilizes the vehicle depending on the selected and scaled parameter set. In this case, by means of the evaluation means, the characteristic measure for the loading state of the vehicle is determined as a function of at least one condition for determining a steady state, wherein the at least one condition a minimum and a maximum yaw rate, a minimum and maximum steering angle, a minimum and maximum speed minimum and maximum yaw rate change and a minimum and maximum steering angle change includes. By means of the evaluation means, a measured value series of measured values collected at regularly successive times is collected, and the characteristic value is determined on the basis of the measured value series, the measured values being checked for plausibility and implausible measured values being filtered out or given a lower weighting.

The invention is based on the object to provide a comparison with the prior art improved method for operating a driver assistance device.

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 a method for operating a driver assistance device of a vehicle, an environment of the vehicle and / or vehicle-related data are detected by means of a plurality of sensors.

According to the invention, in a functional safeguarding of the sensors, scenario-dependent sensor variances and / or sensor variances dependent on the driver assistance device are determined within the framework of error propagation determination, wherein signal variances of the respective sensor are determined from the sensor variances and a safety probability is formed from the sensor variances and signal variances, which are time-variable Threshold is compared, wherein when the threshold value is exceeded, a deactivation of the driver assistance device takes place.

By means of the method according to the invention, it is possible in a particularly advantageous manner to model system-dependent, ie driver-dependent, self-dependent and scenario-dependent sensor variances that change dynamically over time. As a result, an evaluation of the quality of the signals of the sensors at all system interfaces of the driver assistance device based on the time-varying signal variations is possible because each signal of a sensor carries a signal variance that propagates by means of the method according to the invention becomes. In a particularly advantageous manner, it is likewise possible to dynamically evaluate a signal quality of individual sensors with respect to sensor interfaces over time, which is made possible by a comparison of the sensor data with fused data. For evaluation, the time-varying threshold values are used, which are determined from the dynamically variable sensor variances and signal variances of the fused sensor data. Due to the dynamics of the method according to the invention, it is thus possible to safely operate the driver assistance device even in the scenario-related border region of the driver assistance device itself and its sensors. It follows that a balance between the avoidance of false triggering and preservation of an active potential of the driver assistance device can be made more favorable than before.

Compared with static methods known from the prior art, which lead to an artificially binary decision due to the formation of scenario-specific limits of sensors or the entire driver assistance device by the static thresholds, it is possible by means of the inventive method due to the formation of the dynamic thresholds, including flowing transitions display.

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

Showing:

1 schematically a vehicle with a plurality of sensors for detecting an environment of the vehicle and

2 schematically a system architecture of a driver assistance device of the vehicle according to 1 ,

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

In 1 is a vehicle 1 with several sensors 2 to 5 for detecting an environment of the vehicle 1 shown. The sensor 2 is a stereo camera with a detection range E2. The sensor 3 is a long-range radar having a first detection range E3.1 with a medium detection range and a second detection range E3.2 with a wide detection range. The sensors 4 and 5 are each a near-field radar with a detection range E4, E5 with close detection range.

The vehicle 1 includes a driver assistance device 6 , where the sensors 2 to 5 with the driver assistance device 6 are coupled. The driver assistance device 6 is to avoid collisions of the vehicle 1 formed with objects in the environment and not shown in detail or at least to reduce a collision severity, wherein by means of the driver assistance device 6 in particular a in 2 brake shown in detail 7 and steering 8th be controlled.

2 shows a system architecture of the driver assistance device 6 , The driver assistance device 6 are used as input signals via a first interface SS1 sensor data SD2 to SD5 of the sensors 2 to 5 fed.

These are the sensor data SD2 of the sensor designed as a stereo camera 2 especially around pairs of gray pictures.

The sensor data SD3 is radar data of the sensor 3 from the detection range E3.1 with the mean detection range and the second detection range E3.2 with the remote detection range.

In addition, the sensor data formed as radar data SD4, SD5 of the sensor 5 from the detection range E4, E5 supplied with near detection range as input signals.

Furthermore, the driver assistance device 6 as input signals bus signals BS eenes vehicle bus, in particular a CAN bus of the vehicle 1 supplied, wherein the bus signals BS, a longitudinal acceleration, a lateral acceleration and a yaw rate of the vehicle 1 include.

The supplied sensor data SD2 to SD5 and the bus data BD are in a state Z1 in which they are merged by means of a fusion module 9 are suitable.

From the sensor data SD2 to SD5 and the bus data BD are from in the vicinity of the vehicle 1 Detected objects determined relative object data. These relative object data include an object position in the longitudinal and transverse directions of the vehicle 1 for all edges of the respective object, relative velocities and relative accelerations of the objects in the longitudinal and transverse direction, relative yaw rates of the objects as well as various other property measures of the objects, these data as sensor data SD2 to SD5 and bus data BD to the fusion module 9 be supplied.

In this fusion, which takes place in real time, the sensor data SD2 to SD5 are temporally adjusted in a first fusion step FS1, by different sensor latencies 2 to 5 , so-called "sensor delays", are compensated. To this compensation is a relative time shift to the vehicle bus together with the sensor data SD2 to SD5 in the fusion module 9 transfer.

Subsequently, in a second fusion step FS2, scenario-dependent sensor variances SV are determined.

In a third fusion step FS3 is simultaneously based on a Einspurmodels of the vehicle 1 its own movement compensated from the supplied bus data BD.

In a fourth fusion step FS3, the actual fusion and a denoising of the sensor data SD2 to SD5 takes place, the relative object data being in absolute object data OD relative to a reference coordinate system of the vehicle 1 be transformed.

In this case, the disparity is calculated in a so-called semi-global matching algorithm based on two simultaneously acquired gray images. The semi-global matching algorithm is in particular according to Gehrig, S., Rabe, C .: Real-Time Semi-Global Matching to the CPU; In: Proc. ECVW, San Francisco, CA (2010) " executed. With the obtained disparity for all pairs of corresponding pixels, three-dimensional information is determined.

Subsequently, a subset of the three-dimensional information is determined and, in particular, in a so-called six-dimensional view method "Rabe, C., Franke, U., Gehrig, S .: Fast detection of moving objects in complex scenarios; In: IEEE Intell. Veh. Symp., Istanbul (2007) " temporally integrated by means of a multi-dimensional Kalman filter with a second-order process model and thus stabilized and denuded.

The ascertained six-dimensional information, which includes three-dimensional positions and speeds, are grouped into vehicle-like objects, which are subsequently tracked in the gray images on the basis of the sensor data SD3 to SD5, ie the radar data. The grouping of the six-dimensional information on the objects and the tracking of the objects in the images takes place in particular according to "Barth, A., Franke, U.: Estimating the Driving Stages of Oncoming Vehicles From a Moving Platform Using Stereo Vision; In: IEEE Transactions to ITS. (2009) " , In this case, a reliable view-based object detection is possible, in particular up to a distance of 45 m.

The result of the fusion of the sensor data SD2 to SD5 are the absolute object data OD, which include an absolute object position for all edges of the respective object, absolute velocities and accelerations of the objects in the longitudinal and transverse direction, absolute yaw rates of the objects and various other property measures of the objects.

The absolute object data OD are in a fifth fusion step FS5 via a second interface SS2 to a situation analysis module 10 transfer. The supplied object data OD are in a state Z2 in which they are used for situation analysis by means of the situation analysis module 10 are suitable.

In particular, an absolute reference of the object speed and the object acceleration are a prerequisite for a 50 referred to forward simulation in the situation analysis module 10 in an analysis step AS1, wherein the forward simulation in particular according to "Tamke, A., Dang, T., Breuel, G .: A flexible method for criticality assessment in driver assistance systems; In: IEEE Intelligent Vehicles Symposium, Baden Baden (2011) " he follows.

Furthermore, a maneuver classification is carried out by means of the situation analysis module in a second analysis step AS2.

In one by means of the situation analysis module 10 Performed situation analysis based on the maneuver classification and the forward simulation becomes a criticality of a current scenario in the environment of the vehicle 1 assessed, with this possible behavioral options of the vehicle 1 and the identified objects are considered and implemented.

As a result, so-called TTX values TTX (TTX = time to collision / brake / steer / kickdown) in particular after "Tamke, A., Dang, T., Breuel, G .: A flexible method for criticality assessment in driver assistance systems; In: IEEE Intelligent Vehicles Symposium, Baden Baden (2011) " which calculates the basis for the functional decision to operate the driver assistance device 6 form. The TTX values TTX include the time until the imminent collision, a braking, steering and / or acceleration of the vehicle to be carried out 1 as well as other data, such as B. a determined collision position in the environment and a collision angle between the vehicle 1 and the respective object.

The TTX values TTX are sent to a function module via a third interface SS3 11 transfer. The supplied TTX values TTX are in a state Z3 in which they are used to carry out the functional decision by means of the function module 11 are suitable.

The functional decision is whether and in what way by means of the function module 11 the brake 7 and / or the steering 8th and / or passive safety devices 12 of the vehicle 1 For example, airbags, seatbelt pretensioners, active seats and others, are controlled to prevent an imminent collision or at least their consequences for the vehicle 1 and its occupants as well as to reduce the object.

To incorrect triggering the driver assistance device 6 and a resulting risk to the occupants of the vehicle 1 and other road users avoid the sensors 2 to 5 functionally secured. The sensor variances SV are very system-dependent, ie the driver assistance device into which the sensors 2 to 5 are involved influences the quality of the sensor data SD2 to SD5. Furthermore, the sensor variances SV are scenario-dependent, ie depending on the dynamics of a situation, the sensor accuracy can fluctuate greatly in the course of a measurement. Therefore, it is necessary to mitzumodieren these effects mentioned in the functional safety to an error propagation in the driver assistance device 6 to be able to depict. On the basis of the propagated by the system sensor SV variations are in all calculation stages of the driver assistance device 6 Signal variances available.

Therefore, in the functional protection of the sensors 2 to 5 scenario-dependent sensor variations SV and of the driver assistance device 6 self-dependent sensor variances SV determined as part of an error propagation determination, wherein from the sensor variances SV signal variances of the respective sensor 2 to 5 are determined and from the sensor variances SV and signal variances a probability of safety is formed, which is compared with a time-varying threshold, wherein when the threshold value is exceeded, a deactivation of the driver assistance device 6 he follows.

The functional protection of the sensors 2 to 5 is performed in several process sections VA1 to VA3 in a second functional step FFS2.

In a first method section VA1 for securing system-related uncertainties of the driver assistance device 6 becomes a determination of plausibility by means of the sensors 2 to 5 generated signals and the bus data BD, ie the sensor data SD2 to SD5 and bus data BD in the state Z1, the absolute object data OD determined in the state Z2 and the determined TTX values TTX in the state Z3, based on a redundant generation of the signals carried out. In this case, for example, the measured yaw rate is compared with a yaw rate, which consists of a steering angle within the single-track model of the vehicle 1 was determined. Furthermore, the so-called comb-type friction circle is used to check the plausibility of the measured acceleration values. An object orientation, which was determined from data of the stereo camera, is compared with a velocity orientation of the object determined from the radar data, and the object speed determined from the radar data is compared with a demarcation of a radar object distance.

Furthermore, in the first method section VA1, an evaluation of a quality of the signals at all interfaces SS1 to SS3 of the driver assistance device is performed 6 , ie when the sensor data SD2 to SD5 are fed into the fusion module 9 , the object data CD in the situation analysis module 10 and the TTX values TTX into the function module 11 , performed based on expected and dynamically varying signal variances. This leads to novel approaches that allow dynamic reliability control of the signals at the interfaces SS1 to SS3. Expected, dynamically changing signal variances are determined at all interfaces SS1 to SS3. These signal variances are used to calibrate the Kalman filter in the fusion, to perform distance dependent threshold operations to evaluate the quality of the sensor data SD2 to SD5 in state Z1, and to determine a scenario-dependent system uncertainty by aggregating the signal variances.

In order to realize the aforementioned uses of the signal variances, in addition to the signal variances, a transmission of these in the entire driver assistance device is itself possible 6 required. To determine an assumption for the signal variances at the various interfaces SS1 to SS3, the error propagation determination is the sensor variator SV in the entire driver assistance device 6 required.

Two different approaches can be used to determine the sensor variances SV. In a first approach, the sensor variances SV are determined theoretically and in a second approach depending on so-called ground truth data. The ground truth data comprises a set of measured sequences associated with very precise position data, speed data and vehicle acceleration data 1 and the respective object are linked. The required ground truth data will be used in a test environment in which computer-controlled vehicles are objects, in particular as in "Schöner, HP, Hurich, W., Haaf, D .: self-propelled soft crash target for testing assistance systems; In: AAST. (2011) " described, determined.

By directly using the ground truth data for those using the sensors 2 to 5 determined stereo camera objects and radar objects is a reliable modeling of the SV sensor variations possible.

In the next step, the sensor variances SV in all modules of the driver assistance device 6 distributed or deposited, wherein the data transfer via the vehicle bus with a limited resolution and increases the signal variances, which by means of the Kalman Finters within the fusion module 9 be determined.

All determined expected sensor variances SV and signal variances become the function module 11 supplied to determine a probability of likelihood. This probability of security is compared with the time-varying threshold, wherein when the threshold value is exceeded, the deactivation of the driver assistance device 6 he follows. The threshold value is a dynamic distance-dependent threshold value, for which purpose a distance to the respective object is determined from the sensor data SD2 to SD5.

In a first functional step FFS1, a classification of a use case of the driver assistance device is made 6 , a so-called use case classification, performed. That is, from the sensor data SD2 to SD5 and the bus data BD, the scenario in the vicinity of the vehicle 1 determined and depending on the scenario determines which function of the driver assistance device 6 must be activated to avoid a collision or at least to reduce the collision severity.

Depending on the results of the use-case classification, sensor-specific object-related uncertainties and sensor-specific object-related lifetimes are compared with the respectively associated threshold values in a second method section VA2 for securing object-related uncertainties. These threshold values are in particular dynamic range-dependent threshold values.

Also depending on the results of the use-case classification is in a third process section VA3 for the demarcation of applications of the driver assistance device 6 a degree of fulfillment of requirements of the respective application of the driver assistance device 6 , ie a degree of fulfillment of the respective use case determined.

In this case, the degree of fulfillment of the prerequisites in the third method section VA3 is determined by means of a distinction between approaches which involve a control of at least one actuating element of the driver assistance device 6 influence and approaches, which cancel an already activated function of the driver assistance device 6 lead, determined.

Depending on the results of the method sections VA1 to VA3, the safety probability is subsequently determined in a function step FFS3 and compared with the time-variable threshold value. If the threshold is undershot, the function module controls 11 the driver assistance device 6 Depending on the detected use case the brake 7 , the steering 8th and / or the passive safety devices 12 at. An automatic full braking is preferably carried out only when a collision of the vehicle 1 and the object are no longer avoidable. In addition, a control of a drive train of the vehicle 1 possible.

LIST OF REFERENCE NUMBERS

1

vehicle

2 to 5

sensor

6

Driver assistance device

7

brake

8th

steering

9

fusion module

10

Situation Analysis Module

11

function module

12

safety device

AS1, AS2

analysis step

BD

Bus data

E2

detection range

E3.1

detection range

E3.2

detection range

E4

detection range

E5

detection range

FS1 to FS5

fusion step

FFS1 to FFS3

function step

OD

object data

SD2 to SD5

sensor data

881 to SS3

interface

SV

sensor variance

TTX

TTX value

Z1 to Z3

Status

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 102004035578 A1 [0004]

Cited non-patent literature

• Gehrig, S., Rabe, C .: Real-Time Semi-Global Matching to the CPU; In: Proc. ECVW, San Francisco, CA (2010) " [0030]
• "Rabe, C., Franke, U., Gehrig, S .: Fast detection of moving objects in complex scenarios; In: IEEE Intell. Veh. Symp., Istanbul (2007) " [0031]
• "Barth, A., Franke, U.: Estimating the Driving Stages of Oncoming Vehicles From a Moving Platform Using Stereo Vision; In: IEEE Transactions to ITS. (2009) " [0032]
• "Tamke, A., Dang, T., Breuel, G .: A flexible method for criticality assessment in driver assistance systems; In: IEEE Intelligent Vehicles Symposium, Baden Baden (2011) " [0035]
• "Tamke, A., Dang, T., Breuel, G .: A flexible method for criticality assessment in driver assistance systems; In: IEEE Intelligent Vehicles Symposium, Baden Baden (2011) " [0038]
• "Schöner, HP, Hurich, W., Haaf, D .: self-propelled soft crash target for testing assistance systems; In: AAST. (2011) " [0047]

Claims (8)

Method for operating a driver assistance device ( 6 ) of a vehicle ( 1 ), in which by means of several sensors ( 2 to 5 ) an environment of the vehicle ( 1 ) and / or vehicle-related data, characterized in that in a functional protection of the sensors ( 2 to 5 ) scenario-dependent sensor variances (SV) and / or by the driver assistance device ( 6 ) dependent sensor variances (SV) are determined as part of an error propagation determination, wherein from the sensor variances (SV) signal variances of the respective sensor (SV) 2 to 5 ) and from the sensor variances (SV) and signal variances a safety probability is formed, which is compared with a time-variable threshold value, wherein when the threshold value is exceeded a deactivation of the driver assistance device (FIG. 6 ) he follows. Method according to claim 1, characterized in that the functional protection of the sensors ( 2 to 5 ) is carried out in one or more process sections (VA1, VA2, VA3). Method according to Claim 2, characterized in that in a first method section (VA1) for securing system-related uncertainties of the driver assistance device ( 6 ) a determination of a plausibility of by means of the sensors ( 2 to 5 ) signals based on a redundant generation of the signals and an evaluation of a quality of the signals at all interfaces (SS1 to SS3) of the driver assistance device ( 6 ) based on expected and dynamically varying signal variances. Method according to Claim 2 or 3, characterized in that sensor-specific object-related uncertainties and sensor-specific object-related lifetimes are compared with the respectively associated dynamic threshold values in a second method section (VA2) for securing object-related uncertainties. Method according to one of claims 2 to 4, characterized in that in a third process section (VA3) for the demarcation of applications of the driver assistance device ( 6 ) a degree of fulfillment of requirements of the respective application of the driver assistance device ( 6 ) is determined. Method according to Claim 5, characterized in that the degree of fulfillment of the prerequisites in the third method section (VA3) is determined by means of a distinction between approaches which enable an activation of at least one actuating element of the driver assistance device ( 6 ) and approaches which are used to cancel an already activated function of the driver assistance device ( 6 ) is determined. Method according to one of the preceding claims, characterized in that at least two sensor data (SD2 to SD5) of at least one of the sensors ( 2 to 5 ) are merged. Method according to one of the preceding claims, characterized in that as sensors ( 2 to 5 ) at least a stereo camera, a far-field radar and a near-range radar or a subset of these are used.

Images