DE102010045678A1 - Method for monitoring driver condition, involves determining driver condition on basis of track information and comparing track information with deposited samples of track information - Google Patents

Abstract

The method involves determining the driver condition on the basis of track information and comparing the track information with the deposited samples of track information for the determination of the driver condition. The track information is determined from vehicle data, which is determined by an inertial sensor detected sensor data.

Description

The invention relates to a method for monitoring a driver state, in which the driver state is determined on the basis of track information correlated with the driver state, the lane information being compared with stored patterns of lane information for determining the driver state.

It is well-known that the majority of road accidents are primarily due to a misconduct of the driver of the vehicle. If the driver is drowsy, the risk of a collision or near collision is increased many times over.

From the prior art, therefore, various methods for monitoring a driver's condition are known, in which the drowsiness or fatigue of the driver are detected. The drowsiness is detected by an analysis of the current driving style of the driver. To assess sleepiness, lane information is determined which is particularly correlated with drowsiness-impaired driving.

Such methods of assessing sleepiness from lane information which are particularly correlated with drowsiness impaired driving are lacking "F. Friedrichs and B. Yang: Camera-based drowsiness reference for driver classification under real driving conditions; In: Proceedings of the IEEE Intelligent Vehicles Symposium, 2010 and "F. Friedrichs and B. Yang: Drowsiness Monitoring by steering and lane based on driving conditions; In: EUSIPCO, 2010 " known. In this case, a driver monitoring based on a steering behavior, a tracking and a blink behavior is performed.

Out "C. Hasberg and S. Hensel: Online estimation of road map elements using spline curves; In: 12th International Conference on Information Fusion, 2009 " Spline and a Kalman filter are used for online evaluation of train tracks. In "M. Fees: simulation and processing of radar target lists in automobiles; Ph. D. dissertation, Chair of System Theory and Signal Processing, 2008 " a tracking of vehicle destination lists by means of a radar is described. Further describes "M. Miksch, B. Yang and K. Zimmermann: Motion compensation for obstacle detection based on homography and odometric data with virtual camera perspectives; In: Proceedings of the IEEE Conference on Intelligent Vehicles, 2010 the use of a vehicle motion model to determine the vehicle trajectory for motion compensation in video data.

The invention has for its object to provide a comparison with the prior art improved method for monitoring a driver's condition.

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 monitoring a driver state, the driver state is determined on the basis of track information correlated with the driver state, the lane information being compared with stored patterns of lane information for determining the driver state.

According to the invention, the lane information is determined from odometric vehicle data which is determined from sensor data acquired by means of at least one inertial sensor. Generally, the term "lane information" is intended to include the course of movement of the vehicle with respect to the roadway.

Intertial sensors are understood to be sensors by means of which translatory and rotational acceleration forces or motion sequences can be measured. So acceleration sensors are conceivable but also the evaluation of wheel speeds. It is also possible to combine several inertial sensors into a so-called inertial measuring unit, so that the accelerations can be measured up to six degrees of freedom. Assuming a planar movement on a roadway, one can assume less degrees of freedom because the vehicle can only yaw and can not move up or down.

This results in a particularly advantageous manner that a use of optical sensors, in particular costly track cameras or laser scanner (LIDAR), to determine the Track information is not required. Rather, as inertial sensors, in particular already in the vehicle 1 existing sensors of driver assistance devices, preferably sensors, which are provided for the operation of an Electronic Stability Program (ESP for short) used. Thus, on the one hand, the costs for implementing the method for monitoring the driver status are reduced and, on the other hand, the determination of the track information from the odometric vehicle data according to the invention is in contrast to the determination from data acquired by means of a track camera even with missing and incomplete road markings, low incidence of light and at poor visibility possible. Also, inertial sensors are generally distinguished from track cameras by greater robustness to misregistrations, thereby minimizing errors in misalignment detection of track information.

Thus, by means of the method according to the invention, the availability and the recognition rate of lane information are improved without the use of lane cameras for improving the monitoring of the driver's condition.

However, the signals of the inertial sensors could also be used in addition to prepare existing track signals from a camera in order to improve the track warning. Lane changes could be detected reliably even with poor track quality. Overall, the lane data can be improved.

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

Showing:

1 schematically a movement model of a vehicle,

2 schematically two steps for evaluating a linear Kalman filter layer,

3 schematically two steps for advanced evaluation using Kalman filters,

4 schematically by means of track-based sensors and inertial sensors detected lateral positions and

5 schematically a curve of a correlation coefficient between the detected by means of track-based sensors and inertial lateral positions according to 4 ,

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

In 1 is schematically a movement model of a vehicle 1 shown.

The vehicle trajectory includes a lurching between lane markers and a disturbance of the lane or road curvature. Therefore, these bends are removed by a low-pass filter whose cut-off frequency is between the frequency of the roll and the frequency of the strongest curve curvature. The strongest curve curvature depends on the permissible speed. Furthermore, the correlation between track-based and odometric data-based features as well as their connection to drowsiness is compared. An extract of an Attention Assist database with 294 drivers and over 76,000 km was used for this purpose. Many track-based features blend well. The odometry-based zero-crossing rate gives better results than a camera-based determination of the data.

Driver monitoring systems can be controlled with the data determined. These systems use standard sensors to reduce sleepiness related road accidents caused by tired and distracted drivers. To develop and optimize such systems requires a reliable and accurate reference to fatigue or drowsiness. A well-known reference for drowsiness is the subjective self-assessment of a driver according to Karolinska Sleepiness Scale (= KSS), as in Table 1 and "T. Akerstedt, M. Gillberg: Subjective and objective sleepiness in the active individual: In: International Journal of Neuroscience, vol. 52, page 2937, 1980 " shown. Karolinska Sleepiness Scale: Description: 1 Extremely attentive 2 Very attentive 3 attentively 4 Pretty attentive 5 Neither alert nor sleepy 6 Signs of drowsiness 7 Sleepy - Little effort to stay awake 8th Sleepy - some effort to stay awake 9 Sleepy - high effort to stay awake, fights drowsiness Table 1: Karolinska Sleepiness Scale

Out "F. Friedrichs and B. Yang: Camera-based drowsiness reference for driver classification under real driving conditions; In: Proceedings of the IEEE Intelligent Vehicles Symposium, 2010 and "F. Friedrichs and B. Yang: Drowsiness Monitoring by steering and lane based on driving conditions; In: EUSIPCO, 2010 " is an evaluation of features of trace data known which gives trace data that correlate particularly with impaired driving.

In contrast to these known approaches, which use a tracking camera to determine the lane data, odometric vehicle data are determined by means of inertial sensors in the method and its developments, so that the classic track-based features are calculable without the need for a tracking camera. The main assumption behind this approach is that the road curvatures can be evaluated by the odometric data. As in "Clayton: Clothoid, 2006, Available at: http://ww3.cad.de/foren/ubb/uploads/Clayton/Klothoide-Formeln.pdf" described, there is a minimum radius of curvature for each speed limit. For example, the minimum radius of curvature at 120 km / h is 750 meters. In general, this is considered in road construction, in that the road curvature changes only slowly when the road is traveled at high speed.

It can be observed that the lurching in the track takes place at much higher frequencies so that it can be extracted by high pass filters. For the evaluation of the vehicle trajectory, a Kalman filter, an extended Kalman filter and / or a particle filter and optionally a vehicle motion model are used. The use of a Kalman filter is not necessary and there are many other ways to perform a filter function.

In addition to the measured yaw rate (psi_point) and vehicle speed, the GPS position of the vehicle also becomes 1 included. GPS was converted to UTM to use the same metric coordinate system as the coupled odometric position. This is valuable for visualization and system performance evaluation, but is not needed in the online system.

Thus, the track data signals are surely detectable even in the absence of lane markings, bad weather conditions and lighting conditions. Some track-based features; like LANEDEV, ZIGZAGS and others, do not require an absolute position relative to the lane markings. It is therefore sufficient to know the course or the lurching (commuting) within the lane.

Many other aspects of drowsiness detection benefit from this improved vehicle position. Short errors in lane-locking with the camera can be filled in and road condition analyzes benefit from this improved resolution.

To determine the required information, as already explained, test drives were carried out and a database was created with the collected data, covering more than 17,800 real road trips with a driven distance of more than 1.67 million km. After filtering this database for trips over 30 km, with present and plausible drowsiness self-assessment according to the Karolinska Sleepiness Scale, valid tracking data and no measurement errors, remained 76 215 km real trips.

The carrying out of these trips, which in particular also included post-experiments, was carried out in particular according to the method described in "F. Friedrichs and B. Yang: Drowsiness Monitoring by steering and lane based on driving conditions; In: EUSIPCO, 2010 " and "E. Schmidt: Drivers misjudgment of vigilance state during prolonged monotonous daytime driving; In: Accident Analysis & Prevention, vol. 41, pages 1087-1093, 2009 " described method.

In order to obtain the lateral track position signal from the odometric data, the performing steps are presented below. The yaw rate sensor required for the determination has a pattern frequency of z. B. F _S = 1 / T ≈ 50 Hz. The GPS signals were at a sampling rate of z. B. F _S = 1 / T ≈ 1 Hz available. When new GPS data became available, an additional modified Kalman iteration is called every second to update the position according to the GPS data. Thus, the Kalman filter takes over the weighting between intertial sensor-based data and GPS data.

The vehicle movement x shown in the illustrated model is performed according to "M. Fees: simulation and processing of radar target lists in automobiles; In: Ph. D. dissertation; Chair of System Theory and Signal Processing, 2008 " and "M. Miksch, B. Yang, and K. Zimmermann: Motion compensation for obstacle detection based on homography and odometric data with virtual camera perspectives; In: Proceedings of the IEEE Conference on Intelligent Vehicles, 2010 determined and presented as a model.

In this case, the system model according to a position change compensation x (k + 1) = Ax (k) [1] with the position vector x (k) at the moment k and a state transition matrix A.

A second part of the system model provides a measurement equalization z (k + 1) = Hx (k) [2] with the measuring vector z (k) at the moment k. The movement of the vehicle 1 can then be described as

mit der Durchlaufzeit T und dem Lagevektor x(k + 1) gemäß

Where v ⁺ is defined as follows:

with the cycle time T and the position vector x (k + 1) according to

The yaw rate psi_point, d. H. the time derivative of Ψ is measured and is of central importance.

Finally, the following compensation results, which describes the change of position:

verwendet und durch

definiert. Somit ist

Since the system is still non-linear, an extended Kalman filter is needed. For the extended Kalman filter, a non-linear differentiable function f is linearized in each work step, resulting in a current system position x (k). For linearization, the so-called Jacobi matrix becomes the differentiable function

used and through

Are defined. Thus is

Then the measurement matrix H and the vector z (k)

wobei U_e und U_n ostwärts und nordwärts gerichtete UTM-Koordinaten sind. Die Kovarianz-Matrizen Q und R wurden durch den Gebrauch von Hochpass-Maßen ausgewählt.In the event that a new GPS pattern is available, the measurement matrix H and the vector z (k) are expanded according to

where U _e and U _{n are} eastbound and northbound UTM coordinates. The covariance matrices Q and R were selected through the use of high-pass metrics.

To realize an optimal positional evaluation, a Kalman filter is used. The Kalman filter for linear systems is a tool for the evaluation of the position vector, which can be observed by indirect measurements, which are interrupted by a disturbance, as follows: x (k + 1) = Ax (k) + Bu (k) + w (k), [11] z (k + 1) = Hx (k) + v (k). [12]

The model perturbation w (k) and the measurement perturbation v (k) were determined as additional normally distributed white distortion according to: E⌊w (n) w ^T (k) ⌋ = Wδ _nk E [v (n) v ^T (k)] = Vδ _nk p (w) α N (o, Q) p (w) α N (o, R) [13] with an average of zero. E [w (n)] = 0 E [v (n)] = 0 [14]

Model disturbances, measurement errors and initial positions are uncorrelated: E [w (n) v ^T (n)] = 0 E [w (n) z ^T (n)] = 0 E [v (n) z ^T (n)] = 0. [15]

The validity of requirements for the yaw rate Kalman filters, vehicle speed and acceleration sensors is determined according to "M. Fees: simulation and processing of radar target lists in automobiles; In: Ph. D. dissertation; Chair of System Theory and Signal Processing, 2008 " ,

The evaluation of the Kalman linear filter location is calculated in two steps, as in 2 shown. The calculation is carried out as described in " M. Bühren: Simulation and processing of radar target lists in automobiles; In: Ph. D. dissertation; Chair of System Theory and Signal Processing, 2008 " and "G. Welch, G. Bishop: An introduction to the kalman filter; University of North Carolina at Chapel Hill, Department of Computer Science, 2006 " , The steps include a forecast and correction.

The extended Kalman filter is used when the position change is non-linear, as in the present case. The system is described by the non-linear, differentiable functions f and h according to: x (k + 1) = f (x (k), u (k), w (k) = Jx (k) + w (k), [16] z (k + 1) = h (x (k + 1), v (k + 1)). [17]

In the present case, the measurement compensation remains as in equation [12]. The extended Kalman filter attitude assessment will turn, as in 3 shown, calculated in the two steps "forecast" and "correction", but with the linearized function.

A standard GPS sensor was available for the recorded rides. The temporal resolution at 1 Hz is not very high. Also the absolute position is not very accurate. In the "C. Hasberg, S. Hensel: Online estimation of road map elements using spline curves; In: 12th International Conference on Information Fusion, 2009 " and "LVGB: figure and UTM coordinates; State Office for Surveying and Geoinformation Bavaria, 2009 " described UTM representation of GPS data has the advantage that the units use a metric world coordinate system, which is similar to the information obtained by the vehicle data.

As shown in the motion model, the yaw angle φ (k) must be between the vehicle 1 and the track to calculate the lateral distance l (k). The track is evaluated by a low-pass filtered vehicle trajectory using a so-called 2nd-order Butterworth filter with a cut-off frequency of 0.05 Hz. The relative lateral distance Δl to the track is calculated for each sampling time T. One then obtains the lateral distance by updating the weighted lateral position at each sample time: l (k + 1) = l (k) + Δl, [18] with the initial condition I (0) = 0. The lateral distance signal obtained from the vehicle model is again high-pass filtered to eliminate accumulating errors. It is also low-pass filtered to correct interference and road effects with a 2nd-order Butterworth filter with a cut-off frequency of 0.1 Hz.

The extraction of features from the determined track-based data takes place in particular according to "F. Friedrichs and B. Yang: Drowsiness Monitoring by steering and lane based on driving conditions; In: EUSIPCO, 2010 " ,

Table 2 below lists the features selected for the embodiment. The selection is only an example, there are even more features. Track data-based and odometric features have been calculated using the same algorithms. Number: Feature Name: Description: 15 LANEDEV offtrack 17 ZigZags Number of zig zag events when rolling the vehicle 29 LNMNSQ Square lane mean 34 ORA Exceeded range 16 LATPOSZCR Longitudinal position ZCR 30 LNIQR IRQ of the longitudinal position 37 DELTADUR Runtime between the diffraction points 38 DELTALATPOS Mean longitudinal amplitude 39 DELTALATVELMAX Maximum longitudinal speed Table 2: Selection of track-based features

Since lane changes are not determined by the track camera, but by the inertial sensors, the turn signal signal was used to hide intentional lane changes. Three seconds before and ten seconds after the lever operation disappeared. Yaw rates of more than 3 ° / s were also discarded. Furthermore, the system has been defined to be active only at speeds in excess of 80 km / h.

In 4 A comparison of track data-based data-based signals S1 and inertial-data based signals S2 on a physical basis will be explained. The signal S1 is the lateral track position from the spur detection camera or alternative sensors, and signal S2 is the lateral track position signal estimated from inertial data (yaw rate and velocity) using the Inertialdaten basically the signals of other suitable sensors z. As lateral acceleration, wheel speeds or Correvit can be.

The correlation of trace-based and odometric features is shown, as well as the correlation between the odometric features and the Karolinska Sleepiness Scale, the reference to drowsiness, by using the Pearson and Spearman correlation coefficients.

Shown is the lateral difference ("distance") signal obtained after removing the offset. The Spearman correlation coefficients ρ _s between the yaw rate branched features and the original track-based features are shown in Table 3. Number: Feature Name: _p ρ _s 15 LANEDEV 0,006 0.323 17 ZigZags 1,000 0.515 29 LNMNSQ 0.064 0,670 34 ORA 0.443 0.416 16 LATPOSZCR 1,000 0,770 30 LNIQR 0.359 0.693 37 DELTADUR 0.573 0.389 38 DELTALATPOS 0.198 0,429 39 DELTALATVELMAX 0.593 0.536 Table 3: Spearman correlation coefficients and Bravais-Pearson correlation coefficient

wobei cov die Kovarianz und var die Varianz darstellen. Hohe positive oder negative Werte bedeuten eine starke positive oder negative Korrelation, während ein Wert nahe Null ein Zufallsverhältnis anzeigt. Die Spearman Korrelationskoeffizienten arbeiten auf ähnliche Weise, nur mit dem Unterschied, dass eine nicht-lineare Korrelation auch zu hohen Korrelationseffizienten führt. Die Spearman Korrelationskoeffizienten zwischen den gierratenverzweigten Merkmalen und den ursprünglichen spurbasierten Merkmalen werden in Tabelle 4 gezeigt. Nummer: Merkmalsname: ρ_s Lane vs. KSS ρ_s Odometrie vs. KSS 15 LANEDEV 0,211 0,046 17 ZIGZAGS 0,318 0,080 29 LNMNSQ 0,177 0,080 34 ORA 0,325 0,105 16 LATPOSZCR 0,223 0,300 30 LNIQR 0,187 0,100 37 DELTADUR 0,220 0,117 38 DELTALATPOS 0,239 0,079 39 DELTALATVELMAX 0,214 0,106 Tabelle 4: Spearman-Korrelationskoeffizienten The features were calculated and optimized using several methods. In addition to statistical tests, the metrics used were the Bravais-Pearson correlation coefficient ρ _p , the Spearman correlation coefficient ρ _s, and the Fisher metric MDA "M. Welling: Fisher linear discriminant analysis; University of Toronto, Department of Computer Science, Tech. Rep., 2005 " used. The Bravais-Pearson correlation coefficient is calculated to evaluate the linear correlation between the feature F _i and the interpolated filtered Karolinska Sleepiness Scale according to the following equation.

where cov is the covariance and var is the variance. High positive or negative values mean a strong positive or negative correlation, while a value near zero indicates a random ratio. The Spearman correlation coefficients work in a similar way, except that a nonlinear correlation also leads to high correlation coefficients. The Spearman correlation coefficients between the yaw rate-branched features and the original track-based features are shown in Table 4. Number: Feature Name: ρ _s Lane Vs. KSS ρ _s Odometry vs. KSS 15 LANEDEV 0.211 0.046 17 ZigZags 0,318 0,080 29 LNMNSQ 0.177 0,080 34 ORA 0,325 0.105 16 LATPOSZCR 0.223 0,300 30 LNIQR 0,187 0,100 37 DELTADUR 0,220 0,117 38 DELTALATPOS 0.239 0.079 39 DELTALATVELMAX 0.214 0.106 Table 4: Spearman correlation coefficients

5 shows the lane departure feature LANEDEV for a ride. The main correlation between these characteristics can roughly be found in the 5 see.

Some of the features, such as ZIGZAGS, LATPOSZCR and / or the zero-crossing rate correlate particularly with the track-based counterpart. In particular, the feature LATPOSZCR works very well for drowsiness detection.

The main motivation of the present method is to evaluate classical track-based features of inertial sensor-based data rather than camera-based lane data. This has the advantage of providing odometrical data, as an example of inertial-based data, in almost every vehicle nowadays 1 being found. In contrast, tracking cameras are optional equipment and are thus still rarely available in today's fleet. Another major benefit of using inertial sensor-based data is their independence from weather, camera calibration, and the quality of lane marking. This object enormously increases the functionality of the system.

The motion model is for inertial-based sensor signals, and the motion model uses an extended Kalman filter to estimate the lateral track difference with the odometric data. GPS data can also be used to compare and visualize the trace data and the data estimated by inertial sensor. Since the GPS signal is only available every second while the CAN data has a throughput time of several ms, a method for incorporating GPS measurements into the motion model is established. Inertial-based and GPS data are converted to a UTM coordinate system to achieve the same metric representation. The method shows that the features extracted from odometric data correlate particularly with the track-based features, comparing a large amount of data to determine. Generally speaking, some track-based features can approximate very well to odometric data, but others can not. In particular, LATPOSZCR, d. H. the zero-crossing rate of the lateral position works very well due to the continuous system availability of the in-star sensor-based data.

In summary, the method and its developments according to the invention make it possible to evaluate or estimate the relative lateral track position (difference within the track) by using odometric CAN data such as the yaw rate and the speed of the conversion of the wheels. Furthermore, the method comprises mathematical foundations for the evaluation of the vehicle trajectory. Important track-based features can be determined without the need for a track camera.

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 non-patent literature

• "F. Friedrichs and B. Yang: Camera-based drowsiness reference for driver classification under real driving conditions; In: Proceedings of the IEEE Intelligent Vehicles Symposium, 2010 " [0004]
• "F. Friedrichs and B. Yang: Drowsiness Monitoring by steering and lane based on driving conditions; In: EUSIPCO, 2010 " [0004]
• "C. Hasberg and S. Hensel: Online estimation of road map elements using spline curves; In: 12th International Conference on Information Fusion, 2009 " [0005]
• "M. Fees: simulation and processing of radar target lists in automobiles; Ph. D. dissertation, Chair of System Theory and Signal Processing, 2008 " [0005]
• "M. Miksch, B. Yang and K. Zimmermann: Motion compensation for obstacle detection based on homography and odometric data with virtual camera perspectives; In: Proceedings of the IEEE Conference on Intelligent Vehicles, 2010 " [0005]
• "T. Akerstedt, M. Gillberg: Subjective and objective sleepiness in the active individual: In: International Journal of Neuroscience, vol. 52, page 2937, 1980 " [0025]
• "F. Friedrichs and B. Yang: Camera-based drowsiness reference for driver classification under real driving conditions; In: Proceedings of the IEEE Intelligent Vehicles Symposium, 2010 " [0026]
• "F. Friedrichs and B. Yang: Drowsiness Monitoring by steering and lane based on driving conditions; In: EUSIPCO, 2010 " [0026]
• "Clayton: Clothoid, 2006, Available at: http://ww3.cad.de/foren/ubb/uploads/Clayton/Klothoide-Formeln.pdf" [0027]
• "F. Friedrichs and B. Yang: Drowsiness Monitoring by steering and lane based on driving conditions; In: EUSIPCO, 2010 " [0033]
• "E. Schmidt: Drivers misjudgment of vigilance state during prolonged monotonous daytime driving; In: Accident Analysis & Prevention, vol. 41, pages 1087-1093, 2009 " [0033]
• "M. Fees: simulation and processing of radar target lists in automobiles; In: Ph. D. dissertation; Chair of System Theory and Signal Processing, 2008 " [0035]
• "M. Miksch, B. Yang, and K. Zimmermann: Motion compensation for obstacle detection based on homography and odometric data with virtual camera perspectives; In: Proceedings of the IEEE Conference on Intelligent Vehicles, 2010 " [0035]
• "M. Fees: simulation and processing of radar target lists in automobiles; In: Ph. D. dissertation; Chair of System Theory and Signal Processing, 2008 " [0047]
• M. Bühren: Simulation and processing of radar target lists in automobiles; In: Ph. D. dissertation; Chair of System Theory and Signal Processing, 2008 " [0048]
• "G. Welch, G. Bishop: An introduction to the kalman filter; University of North Carolina at Chapel Hill, Department of Computer Science, 2006 " [0048]
• "C. Hasberg, S. Hensel: Online estimation of road map elements using spline curves; In: 12th International Conference on Information Fusion, 2009 " [0051]
• "LVGB: figure and UTM coordinates; State Office for Surveying and Geoinformation Bavaria, 2009 " [0051]
• "F. Friedrichs and B. Yang: Drowsiness Monitoring by steering and lane based on driving conditions; In: EUSIPCO, 2010 " [0053]
• "M. Welling: Fisher linear discriminant analysis; University of Toronto, Department of Computer Science, Tech. Rep., 2005 " [0059]

Claims (6)

A method for monitoring a driver state in which the driver state is determined on the basis of correlated with the driver state lane information, wherein for determining the driver state, the lane information is compared with stored patterns of lane information, characterized in that the lane information from odometric vehicle data are determined, which by means of At least one inertial sensor detected sensor data are determined. A method according to claim 1, characterized in that by means of the at least one inertial sensor as sensor data, a yaw rate, a yaw acceleration, a yaw angle (φ (k), eine), a longitudinal acceleration, a lateral acceleration, a steering wheel angle and / or a wheel speed of at least one wheel of the Vehicle are detected. A method according to claim 1 or 2, characterized in that as lane information, a lane departure of the vehicle, a commuting of the vehicle in a lane, a lane keeping frequency and / or zig-zag travels are determined. Method according to one of the preceding claims, characterized in that the odometric vehicle data are determined from the sensor data based on calculations of clothoids, circular arcs and / or splines. Method according to one of the preceding claims, characterized in that one or more filters are used in the determination of the odometric vehicle data from the sensor data. Method according to one of the preceding claims, characterized in that in the monitoring of the driver state from the lane information a wakefulness of the driver is determined.

Images