METHOD FOR IDENTIFYING THE DRIVER OF A VEHICLE

Description

[0001] The invention relates to a method for identifying the driver of a vehicle based on characteristic or state parameters of the vehicle. State of the art

[0002] From DE 10 2016 218 719 A1, a system for identifying the driver of a passenger car or truck based on operating patterns is known. The operating patterns are recorded using an accelerometer integrated into a control element operated by the driver. To identify the driver, the recorded operating pattern is compared with stored operating patterns.

[0003] US 10 417 838 B2 describes a system for classifying driving maneuvers such as turns, roundabout maneuvers, and braking maneuvers. The classification is based on various vehicle-related measurements, such as vehicle speed, acceleration, and GPS positioning.

[0004] US Regulation 2010 / 0209890 A1 describes the classification of driving proficiency levels using neural networks.

[0005] From CN 111 738 337 A it is known to determine the distraction state of a driver in various traffic situations based on eye movements and driving data using a long-short-term memory network.

[0006] A method for identifying the driver of a vehicle is known from EP 2 261 096. Disclosure of the invention

[0007] The method according to claim 1 can be used to identify the driver of a vehicle, in particular a motorized vehicle. The vehicle may be a single-track vehicle such as a motorcycle or a scooter, or a two-track vehicle, in particular a passenger car or a truck.

[0008] Using the method according to the invention, the current driver of a vehicle can be identified by comparison with one or more drivers from a reference pool. Both a positive identification is possible if the current driver matches one of the drivers from the reference pool, and a negative identification is possible if the current driver does not match any driver from the reference pool.

[0009] The process takes place in two steps: first, during a training phase, driver-specific parameters or state variables are determined for one or more drivers, forming a reference pool. The training phase occurs while the vehicle is being driven. During this phase, the current parameters or state variables in the vehicle, which change in response to driver input, are determined, primarily using sensors installed in the vehicle.

[0010] Using a classification model, driver-specific classification parameters are determined from the characteristic or state variables.

[0011] The parameters under consideration are primarily driver-specific variables such as transmission shift points, brake pressures, and the like. The state variables considered, which are determined via the vehicle's own sensors, include, for example, translational and rotational state variables in or around the vehicle's longitudinal, lateral, and vertical directions at the position, velocity, and / or acceleration levels.

[0012] In a classification phase following the training phase, the trained classification model is fed with driver-specific parameters or state variables, which are determined in the vehicle during a drive with a current driver. Relatively short subsequences of the parameters or state variables are used for this purpose, and these subsequences are typically shorter than the training sequences used to train the classification model for one or more drivers during the training phase.

[0013] The training sequences of the training phase typically have a duration of, for example, several minutes, whereas the subsequences from the classification phase are considerably shorter than the training sequences, for example, only a maximum of one-tenth of their length, and typically only a few seconds, for example, a maximum of 10 seconds. The various subsequences can be completely different from one another, for example, by considering only subsequences from different time periods, or by using subsequences that overlap in time. The time span covered by all the considered subsequences is generally shorter than the time span of the training sequence.

[0014] The training phase only needs to be completed once to establish the reference pool. The classification phase can be repeated as often as necessary, particularly after each vehicle restart to identify the current driver. It is also possible to repeat the training phase to expand the reference pool with additional drivers. Furthermore, the training phase can be repeated to improve parameter classification after the current driver has been identified, potentially for the entire duration of the current driving time with that driver.

[0015] During the classification phase, driver-specific characteristics or parameters are determined across several subsequences. The trained classification model is fed each subsequence, from which a probability of match between that subsequence and a training sequence of a driver from the reference pool is calculated. The probability of match across multiple subsequences can then be aggregated into an overall probability, which is used to identify the current driver from the reference pool.

[0016] This approach has the advantage of achieving a high classification rate with high accuracy. Classification can be carried out quickly without compromising the classification rate. During the classification phase, the results are accumulated, each relating to a specific subsequence. This allows for rapid and accurate driver identification.

[0017] The probability of a match for each subsequence can, according to an advantageous implementation, be determined as a function of the characteristic or state variables, where the function represents the classification model. The current characteristic or state variables of the subsequence under consideration are fed into the classification model, which yields a probability of a match for that subsequence. This process is performed for several different subsequences with characteristic or state variables, with a probability of a match being determined for each subsequence. From the accumulation of these probability of a match, the overall probability by which the current driver is identified can be calculated.

[0018] Several mathematical relationships can be considered for the classification model. In one advantageous implementation, the classification model is designed as a neural network. Since the input parameters or state variables are temporally ordered, a recurrent network, such as a long-short-term memory network (LSTM), can be used as the classification model. Other classification methods, such as a k-nearest neighbors algorithm, are also possible.

[0019] According to a further advantageous embodiment, the classification model outputs a discrete probability for each subsequence, and all these discrete probabilities are combined to form an overall probability. This is achieved, in particular, by adding the discrete probabilities assigned to the subsequences. The result is a probability value for each driver in the reference pool, such that the current driver corresponds to the respective driver in the reference pool. The current driver is assigned to the driver in the reference pool with the highest probability.

[0020] After identifying or failing to identify the current driver, various measures can be taken. If the classification shows that the current driver has no match in the reference pool, an immobilizer can be activated in the vehicle, for example. If the vehicle is currently in motion, a warning signal can be generated, which is displayed in the vehicle and / or transmitted externally, for example, to a reporting center. In the case of an autonomous or semi-autonomous vehicle, it can be brought to a stop automatically.

[0021] If the driver classification results in a clear assignment to a driver from the reference pool, driver-specific settings can be made in the vehicle. This includes, for example, individually adjusting parameter settings in driver assistance systems, engine settings, transmission settings, or chassis settings. If the current driver does not match any of the drivers in the reference pool, standard settings can be used.

[0022] The invention further relates to a control unit containing means configured to carry out the method described above. The means comprise at least one storage unit, at least one processing unit, a control unit input, and a control unit output. Vehicle characteristic or state variables can be supplied to the control unit as input variables. An identification signal can be generated in the control unit that identifies the current driver from the reference pool. Systems in the vehicle can be controlled via the control unit, for example, the immobilizer, the driver assistance systems, the engine, the transmission, or the chassis.

[0023] The invention also relates to a vehicle equipped with a control unit as described above. The vehicle is either a single-track vehicle, for example a motorcycle, or a two-track vehicle, for example a passenger car or a truck.

[0024] The invention further relates to a computer program product comprising program code designed to execute the aforementioned process steps. The computer program product runs in the aforementioned control unit.

[0025] Further advantages and practical designs can be found in the additional requirements, the figure description, and the drawings. These show: Fig. 1 a motorcycle equipped with a control unit for driver identification, Fig. 2 a flowchart with process steps for driver identification.

[0026] In Fig. 1 Figure 1 schematically depicts a single-track vehicle designed as a motorcycle 1. The motorcycle 1 is equipped, in a manner known per se, with a hydraulic braking system 2, comprising a front brake unit on the front wheel, a rear brake unit on the rear wheel, a brake control unit, and a hydraulic pump for adjusting the hydraulic brake pressure. The brake control unit incorporates an anti-lock braking system (ABS), which functions as a driver assistance system.

[0027] Motorcycle 1 is further equipped with a drive motor to which an engine control unit 3 is assigned. Motorcycle 1 is also equipped with sensors, including an inertial sensor 4 and speed sensors 5 and 6 on the front and rear wheels, respectively. The sensor data from sensors 4, 5, and 6 are fed as input signals to an evaluation unit 7 on the engine control unit 3, where rider classification can be performed. The evaluation unit 7 can be part of the engine control unit 3 or a separate control unit.

[0028] The wheel speed of the front and rear wheels is measured via the speed sensors 5 and 6. Driving parameters of the motorcycle 1 can be measured via the inertial sensors 4, including the vehicle speed, longitudinal and lateral acceleration, and the yaw rates about the vehicle's longitudinal axis, transverse axis, and vertical axis.

[0029] In Fig. 2 The diagram shows a flowchart with procedural steps for identifying the current rider on the motorcycle.

[0030] A prerequisite for identification is the existence of a reference pool containing classification parameters for at least one, preferably several, different drivers. The reference pool is created in advance using the drivers' driving data. This is achieved by determining training sequences with vehicle characteristics or state variables for each driver in the reference pool during their driving. Driver-specific classification parameters are then determined from a classification model f(x), which is trained using these training sequences of the vehicle characteristics or state variables x, and stored. The vehicle characteristics or state variables are determined using the vehicle's own sensors and include, for example, transmission shift points, brake pressures, and speed and acceleration values.

[0031] The classification model f(x) with the characteristic or state variables x is preferably designed as a neural network, in particular as a recurrent network such as a Long Short-Term Memory Network (LSTM).

[0032] The reference pool can be continuously expanded to include additional drivers by training the classification model f(x) with the training sequences of these drivers, consisting of vehicle characteristics or state variables.

[0033] In a first process step V1, a sequence of length l2 containing characteristic or state variables of the current driver in the vehicle is acquired using sensors. The length l2 of the sequence refers to the number of data points in this sequence and is generally shorter than the training sequences. In a subsequent process step V2, the sequence of length l2 is divided into several subsequences of length l1, resulting in a total of l2 / l1 subsequences.

[0034] In the next process step V3, each subsequence of length l 1 with the characteristic or state variables x is entered into the classification model f(x), which in the following process step V4 provides a discrete probability value P driver , subsequence for the subsequence in question: P driver , subsequence = f x

[0035] This yields the same number of probability values ​​Pdriver,subsequence as the number l2 / l1 of subsequences. The highest probability values ​​that can be assigned to a driver from the reference pool are considered.

[0036] In the next step V5, an overall probability P driver, sequence is formed by adding the discrete probabilities P driver, subsequence for each subsequence: P driver sequence = ∑ n = 1 n = l 2 / l 1 P driver subsequence

[0037] From the total probability P driver, sequence, in process step V6 the driver can be determined. driverfrom the reference pool, the driver with the highest value of the total probability P driver, sequence is identified: driver = argmax driver P driver sequence

[0038] It may be useful to specify a probability threshold for identifying a driver, which must be exceeded for unambiguous identification. This prevents misidentifications in cases where the current driver does not match any driver in the reference pool (non-identification).

[0039] Depending on whether the current driver is identified or not, various measures can be taken. If the current driver has no match in the reference pool, an immobilizer can be activated in the vehicle. If the vehicle is currently in motion, a warning signal can be generated and / or the vehicle can be brought to a stop automatically.

[0040] If a driver can be clearly identified from the reference pool, driver-specific settings can be applied to the vehicle, such as parameter settings in driver assistance systems, engine settings, transmission settings, or chassis settings. If the current driver cannot be identified, default settings can be used.

Claims

1. Computer-implemented method for identifying the driver of a vehicle based on characteristic or state variables of the vehicle, wherein - in a training phase for one or more drivers forming a reference pool, driver-specific characteristic or state variables (x) of the vehicle are determined during a journey and driver-specific classification parameters of a classification model (f(x)) having training sequences of the characteristic or state variables (x) are determined from the classification model (f(x)), - in a subsequent classification phase, driver-specific characteristic or state variables (x) are determined during a journey and a plurality of subsequences of the characteristic or state variables (x) are supplied to the classification model (f(x)) and a match probability (Pdriver, subsequence) is determined for the match between the sequence and the training sequence, wherein the match probabilities (Pdriver, subsequence) of a plurality of subsequences are combined to form a total probability (Pdriver, sequence) that is used to identify the current driver from the reference pool, - wherein the classification model (f(x)) is designed as a neural network.

2. Method according to Claim 1, characterized in that the match probability (Pdriver, subsequence) is associated with the classification model (f(x)) according to P driver , subsequence = f x , where Pdriver, subsequence means the match probability f(x) means the classification model X means the characteristic or state variables.

3. Method according to either of Claims 1 and 2, characterized in that the classification model (f(x)) is designed as a recurrent network, for example as a long short-term memory network (LSTM).

4. Method according to one of Claims 1 to 3, characterized in that there is a discrete probability (Pdriver, subsequence) at the output of the classification model (f(x)) for each subsequence, which probabilities are combined to form the total probability (Pdriver, sequence).

5. Method according to Claim 4, characterized in that the total probability (Pdriver, sequence) is formed by adding the discrete probabilities (Pdriver, subsequence) for each subsequence: P driver , sequence = ∑ n = 12 / l 1 n = 1 P driver , subsequence 6. Method according to one of Claims 1 to 5, characterized in that that driver from the reference pool with whom the current driver has the highest value of the total probability (Pdriver, sequence) is identified: driver − argmax driver P driver , sequence 7. Method according to one of Claims 1 to 6, characterized in that, if the current driver does not match a driver from the reference pool, an immobilizer in the vehicle is activated.

8. Method according to one of Claims 1 to 7, characterized in that, after the driver has been identified, driver-specific settings are made in the vehicle.

9. Control unit in a vehicle for carrying out the method according to one of Claims 1 to 8, wherein characteristic or state variables (x) of the vehicle can be supplied to the control unit (3, 7) as input variables and a signal which identifies the current driver from the reference pool can be generated in the control unit (3, 7).

10. Vehicle having a control unit according to Claim 9.

11. Computer program product comprising program code designed to carry out steps of the method according to one of Claims 1 to 8 when the computer program product runs in a control unit (3, 7) according to Claim 9.