Computer-implemented method for determining a driver's emergency situation

Abstract

TITLE: Computer-implemented method for determining an emergency situation of a driver Computer-implemented method for determining an emergency level of a driver (12) in his own vehicle (10) based on contextual information data (26), comprising: a) inputting vehicle dynamic data (24) representing at least one quantity related to the movement of the own vehicle (10), b) inputting the contextual information data (26) representing at least one property external to the own vehicle (10), c) generating a feature vector of the vehicle dynamic data (24) and the contextual information data (26), d) classifying the feature vector into an emergency class or a non-emergency class and tracking the quantity of occurrences of the emergency class with an anomaly counter, and e) determining an emergency level based on the value of the anomaly counter. Figure 1

Description

Title of the invention: Computer-applied method for determining an emergency situation of a driver FIELD OF THE INVENTION

[0001] The present invention relates to a computer-implemented method for determining an emergency situation of a driver. STATE OF THE ART

[0002] Document US 10,748,419 B1 describes a method for generating a vehicle-to-vehicle traffic alert and updating the vehicle usage profile. Various elements are detecting by one or more processors associated with a first vehicle, an abnormal traffic condition in the operational environment of the first vehicle. A generated electronic message is transmitted by radio, by a transceiver installed in a vehicle associated with the first vehicle to alert a neighboring vehicle of the abnormal traffic condition and to enable the neighboring vehicle to avoid the abnormal traffic. The first vehicle receives telematics data regarding the operation of the neighboring vehicle after the neighboring vehicle receives the electronic message and transmits the telematics data to a remote server to update a profile.

[0003] Document US 2022 / 0 126 840 A1 describes a method for augmenting transportation services using event detection. The method comprises collecting first sensor data generated by different sensors associated with a set of vehicles. The first sensor data comprises sensor outputs indicating a set of reckless driving events. The sensor outputs are augmented by angular rotation to obtain augmented sensor outputs. A prediction model is trained on the augmented sensor outputs. Target sensor data associated with a target vehicle is provided as input to the training prediction model and based on an output of the trained prediction model the occurrence of a reckless driving event is detected in real time or near real time.Based on the count of reckless driving events associated with the target driver in a cumulative driving distance, the target driver's level is determined.

[0004] Document AU 2021 102 962 A4 describes a driver behavior analysis system based on mood detection. Image-based facial expression classification by a convolutional neural network is used to control the driver. An emotion recognition system, integrated in real time, makes it possible to detect, exploit and evaluate the emotional state of the driver.

[0005] Documents IN 2019 21 028 280 A and IN 2018 41 018 436 A describe methods for detecting reckless driving and excessive speed.

[0006] Document DE 10 2015 215 400 A1 describes a method for providing an emergency measure for situational forecasting of environmental information to a motor vehicle driver. The emergency measure indicates the probability that the driver of the motor vehicle is in a hurry.

[0007] PURPOSE OF THE INVENTION

[0008] The present invention aims to improve emergency detection during vehicle operation.

[0009] DISCLOSURE AND ADVANTAGES OF THE INVENTION

[0010] For this purpose, the invention relates to a computer-applied method for determining a level of urgency of a driver in his own vehicle based on contextual information data,

[0011] the method consisting of:

[0012] a) capture the dynamic vehicle data representing at least one quantity linked to the movement of the vehicle itself,

[0013] b) capturing contextual information data representing at least one property external to the vehicle itself,

[0014] c) generating a feature vector of its vehicle dynamic data and contextual information data,

[0015] d) classifying this feature vector into an emergency class or a non-emergency class and tracking the quantity of occurrences of the emergency class with an anomaly counter, and

[0016] e) determining an urgency level based on the value of the anomaly counter.

[0017] The invention considers contextual information in addition to other sources of information. The classification is done by supervised or unsupervised or semi-supervised classification algorithms. Reckless driving is treated as an indicator of a driver emergency. By taking into account the contextual information, different causes of abnormal driving behavior can be determined more reliably. The determining contextual information includes, but is not limited to, road congestion, road conditions, lane changes and road-cutting behaviors by neighboring drivers. Since the contextual information characteristically influences the driver's driving style, this influences the classification accuracy of the emergency detection model.The described method allows to detect the driver's urgency with greater precision and / or greater reliability by taking into account contextual information.

[0018] Preferably, in state a) the vehicle dynamic data comprises any one of the data: speed of the own vehicle, acceleration of the own vehicle and jerk of the own vehicle. The speed of the own vehicle, acceleration and / or jerk are characteristic indications of driving urgency. Generally, it is accepted that high speeds, accelerations and jerks are associated with a high degree of urgency. Consequently, taking into account at least one of these quantities allows a more precise and / or more reliable determination of the overall urgency.

[0019] Preferably, in step b), at least one property is chosen from any one of the following properties: weather data, congestion data, road condition data, traffic data and vehicle dynamics data of at least one other vehicle. This contextual information data makes it possible to mitigate or have a better understanding of the driving state. It may happen, for example, that the congestion or the road condition forces the driver of the own vehicle to slow down even though he is actually in an emergency. The vehicle dynamics data of the other vehicles can also increase the understanding of the general current driving state if the own vehicle moves on average faster or slower than the other vehicles in the same conditions.

[0020] Taking into account all these different (and more or less unindicated) circumstances allows a more precise and more reliable determination of the overall emergency.

[0021] Preferably, in step c), the feature vector is generated to include any of the following: maximum speed, maximum speed deviation Avavg, vehicle jerk range, maximum acceleration, maximum jerk, geometric mean of speed deviation Avavg, maximum acceleration deviation AJavg, third quartile of speed deviation Avagb, maximum jerk deviation Avavg, standard deviation of acceleration, standard deviation of speed, third quartile of acceleration deviation, average acceleration, speed deviation range Avavg, acceleration deviation range and at least one statistical feature extracted from vehicle dynamic data of the own vehicle and / or at least one other vehicle. Simulation by the applicant has shown that these quantities are to a very high extent correlated with urgency.As a result, taking more closely correlated quantities increases the accuracy and / or reliability of determining overall urgency.

[0022] Preferably, the statistical characteristic comprises one of the following information: speed distribution, acceleration distribution, jolt distribution, maximum speed distribution, speed deviation distribution. maximum Avavg, vehicle jerk distribution range, maximum acceleration distribution, maximum jerk distribution, geometric mean of speed deviation distribution Avavg, maximum acceleration deviation distribution AJavg, third quartile speed deviation distribution Avavg, maximum jerk deviation distribution Avavg, acceleration standard deviation distribution, speed standard deviation distribution, third quartile acceleration deviation distribution, average acceleration distribution, speed deviation range distribution Avavg, acceleration deviation range distribution. Taking into account statistical characteristics allows a better view of the emergency behavior of the driver of the own vehicle.It is possible for example that a particular driver of the own vehicle generally drives faster with more acceleration and / or more jerk than another driver of the own vehicle. Considering the statistical distribution this allows to determine the overall urgency with greater accuracy and / or reliability than the effects of baseline differences in driving styles of different drivers according to the mixed urgency level.

[0023] Preferably, in step d), the classification is done with the supervised binary classification method or the multi-class classification method, generally a random forest approach or an unsupervised classification method such as the kernel density estimation method to classify the feature vector into an emergency class or a non-emergency class. The classification into two classes makes it possible to improve the certainty of the attribution. As a result, one can always have a definite determination for the driver in an emergency or a non-emergency situation.

[0024] Preferably, the binary classification method and the multi-class classification method provide a result indicating a normal state or a result indicating urgency, wherein the multi-class classification method further provides at least one result indicating another class of driver behavior. It is also possible to include other driver states in the evaluation which may or may not be related to urgency or driving style such as "reckless", "calm" etc.

[0025] Preferably, the classification method comprises a random forest approach or an XGBoost algorithm approach. The XGBoost algorithm is described in the document "T. Chen, C. Guestrin, XGBoost: A Scalade Tree Boosting System". This algorithm or method is available under arXiv:1603.02754v3; the expression XGBoost represents eXtreme Gradient Boosting which is a very efficient and very versatile algorithm for machine learning, in particular for tabular structured data. The XGBoost method is part of a family of algorithms for Boosting. Boosting is a set of techniques that add new models to correct errors made by existing models. Models are added sequentially until there is no significant improvement. Specifically, the XGBoost method uses Gradient Boosting, which means it uses the gradient descent algorithm to minimize losses when adding new models. This method improves three things: the loss function is optimized, training is easier to enable predictions, and the additive model to add weak learners to minimize the loss function. In particular, the XGBoost method can automatically handle missing data. When it encounters a missing value, it assigns a gradient descent direction for the missing values ​​according to the direction that most reduces the loss function.Another feature of the XGBoost method is regularization, which prevents overfitting. This is an advantage over traditional implementations. It features L1 (lasso regression) and L2 (ridge regression). The XGBoost method uses a depth-first approach and prunes trees backward. Unlike the traditional method where trees grow to their full depth and are then pruned, the XGBoost method stops growing the tree when it encounters a negative loss in the cut. The XGBoost method is preferably used to efficiently handle sparse data, which is common in many real-world datasets, especially in contexts such as classification.

[0026] Preferably in step d) classification is carried out using the unsupervised classification method. Preferably, the classification method comprises the kernel density estimation method for classifying the feature vector into an emergency class or a non-emergency class. The kernel density estimation provides a better, more precise and reliable understanding than other supervised classification methods. Since the computational effort is relatively low, it is particularly suitable for integrated applications.

[0027] Preferably, in step d) the emergency level is determined by shifting the average approach which consists of dividing the anomaly counter into a current time step which is measured from the start of a trip with the own vehicle. Thus, more weight is given to anomalies which may occur in a short time window. In other words, if the driver occasionally exhibits emergency behavior, it is less likely for the driver to be in an emergency, whereas if the driver exhibits multiple occasions of emergency behavior in a short time interval, it is more likely that the driver is in an emergency. As a result, the accuracy and / or reliability of the method is improved.

[0028] Preferably, step e) comprises comparing the urgency level to the historical urgency level data and determining that the driver is in an emergency if the urgency level exceeds a predefined threshold, preferably given by the historical urgency level. Thus, it is possible to take into account different driving styles of the driver of the own vehicle by comparing the current driving to his own historical driving style. Different drivers typically have different driving styles so that a driving style can be considered more urgent for one driver but which will be considered as the normal style for another. Consequently, the comparison between the historical driving urgency level and the current level allows a more precise and / or more reliable determination of the overall urgency.

[0029] The invention also relates to a computer-implemented method for managing a vehicle driving assistance system, this method consisting of:

[0030] - apply the method described above to determine the level of urgency in comparing the emergency level to historical emergency level data and determining that the driver is in an emergency if the emergency level exceeds a predefined threshold given by the historical emergency level;

[0031] - generate a control signal to display a warning message by the vehicle to increase braking power, to increase the sensitivity of the collision sensors and / or to decrease the maximum speed and / or acceleration of the vehicle. The method is preferably applied to a driver assistance system which can thus adapt the behavior of the vehicle to the urgency of the driver of the own vehicle.

[0032] The method provides a vehicle driving assistance system comprising means for applying the method described above.

[0033] The method also relates to a computer program with instructions which, when the program is executed by a computer, cause the computer to apply the method or if the program is executed by the driver assistance system, the driver will execute the method.

[0034] The subject of the invention is a computer-readable medium or a data-carrying signal comprising the computer program.

[0035] Unlike known approaches that classify the driver's driving style into categories such as "normal" "reckless" "calm" etc. using the historical driving styles of different drivers. The present invention considers other information such as contextual information. In the solution presented here, the classification can be done by supervised, unsupervised or semi-supervised classification algorithms. Reckless driving can be treated as an indicator of an emergency driver.

[0036] Consideration of contextual information makes it possible to evaluate abnormal driving behavior. The determining factors include, but are not limited to, road congestion, road condition, lane changes and the cutting behavior of neighboring drivers etc. These elements characteristically influence the classification accuracy of the emergency detection models. In the described method, the improved method is used to detect the driver's emergency if the contextual information is incorporated for the classification.

[0037] In general, the method described comprises capturing contextual information. It is preferable to have different sources for contextual information. A source is on board which makes it possible to provide dynamic vehicle data such as vehicle speed, vehicle jolts or vehicle acceleration.

[0038] Vehicle sensors are another source that monitor the surrounding area. The sensor is a camera, ultrasonic sensors or similar sensors to capture lane changes or obstacles (including other vehicles, pedestrians or others). Another source is weather data. Weather data is provided via the internet, for example from third-party weather information sites.

[0039] Another source is a sensor monitoring the road condition. Alternatively or additionally, the road condition may be obtained from the traffic information service. For example, the road condition may have different classes such as bumpy, wet, slippery, curved, low traffic, highway, speed limit, etc.

[0040] Another source is traffic data obtained via the internet, which can indicate a traffic jam.

[0041] The next step is the classification of the driver's driving style. The driving style classification is preferably done to classify the driver as being in an "emergency class" (often also referred to as "abnormal" driver behavior) or a "non-emergency" class (sometimes referred to as "normal" driver behavior), using a trained classifier and the contextual information gathered in the previous step. The trained classifier may be trained in a simulated environment in which test subjects are given different driving situations or tasks and the level manually labeled. A supervised classification method is most appropriate using a random forest decision. A suitable unsupervised classification method is kernel density estimation (KDE).

[0042] After classification, the recent results of the classification are evaluated. The cases of abnormal behavior are aggregated over time, preferably with a moving average over time used to determine the level of urgency. This level Emergency level - also called current emergency level - is used to determine whether an emergency pattern persists. The current emergency level is then compared to the driver's historical driving style. Using a reference distribution of the emergency level, the system determines whether the driver is behaving differently compared to historical data, i.e., whether the driver is driving more recklessly than usual.

[0043] If the level of reckless driving exceeds a certain threshold, the driver is classified as being in an emergency.

[0044] In other words, the emergency pattern is typically determined by persistence, if a threshold is exceeded by the emergency level (this may be based on the historical emergency data of the respective driver).

[0045] As an application, if it has been determined that the driver is in an emergency, a control signal can be generated that causes the driver assistance system to perform different functions. For example, the driver can be informed by displaying a warning message. Another possibility for the driver assistance system is to limit the maximum speed or maximum acceleration of the vehicle. Another possibility is to increase the braking power of the vehicle. Another idea is to set the collision warning system to a more sensitive setting, i.e., an earlier warning and / or a possible braking procedure. If this is possible, the data collection range by the vehicle sensors is increased.

[0046] Overall, road safety can be increased by determining whether the driver is in an emergency and in response, determining that the driver should be in an emergency to activate various safety measures designed to reduce the risk of collision or other event related to the driver in an emergency. Brief description of the drawings

[0047] The present invention will be described below in more detail with the aid of embodiments shown in the accompanying drawings in which:

[0048] [Fig. 1] represents a state of traffic,

[0049] [Fig.2] represents a data stream,

[0050] [Fig.3] represents a diagram of a classification method,

[0051] [Fig.4] represents a diagram of a classification method,

[0052] [Fig.5] represents an embodiment of a method for determining a measurement emergency.

[0053] DESCRIPTION OF AN EMBODIMENT

[0054] [Fig.l] shows a traffic situation. A driver 12 ([Fig.2]) is driving his own vehicle 10 on a road 14. Other vehicles 16 are present. The driver 12 drives his own vehicle 10 from a starting point to a destination. The own vehicle 10 has a driving assistance system 18. The driving assistance system 18 is shown separately from the own vehicle 10 for reasons of clarity.

[0055] The driving assistance system 18 is coupled to one or more sensors 20 installed in the vehicle 10 itself. The sensors 20 are configured as ultrasonic sensors, image sensors in general, camera, lidar sensor, radar sensor, accelerometer, tachometer or similar sensor.

[0056] The other vehicles 16 have the possibility of communicating with the driving assistance system 18 via a telecommunications link 22. The telecommunications link 22 is typically configured to encompass other vehicles 16 which are near the own vehicle 10.

[0057] The driver assistance system 18 can be connected to the internet for other particular sources of data such as weather data, traffic data (traffic jams), traffic data and other road condition data.

[0058] According to [Fig. 2], the driver assistance system 18 receives the vehicle dynamic data 24 and the contextual information data 26. The vehicle dynamic data 24 comprises quantities related to the movement of the own vehicle 10 and if possible, to that of other vehicles 16 such as speed, acceleration and / or jolts. The vehicle dynamic data 24 is obtained for a time period of one minute with, for example, one data point every second. In this example, the vehicle dynamic data 24 comprises a vector of length 60. It should be noted that the sampling period and / or interval, i.e. the data points per second, can be modified appropriately.

[0059] From such quantities, other dynamic vehicle data 24 are formed such as the average speed Vavg, the average road speed Vavg_road, the average shock Javg and the average road shock Javg_road.

[0060] The contextual information data 26 is the representation of quantities describing the environment or the environment zone of the own vehicle 10. Examples of contextual information data 26 include, but are not limited to, weather data, traffic jam data, traffic data and road condition data. The sensor data provided by the sensors 20, in particular the image sensors which monitor the environment of the own vehicle 10, are also part of the contextual information data 26.

[0061] The driver assistance system 18 contains the record of the driver profile 28 of each driver 12. The driver profile 28 includes the historical emergency data 30 representing the past driving style of the driver. 12 as related to the emergency and serving as a reference to determine whether driver 12 is in an emergency.

[0062] Two classification concepts will be described in more detail with reference to [Fig. 3] and [Fig. 4]. The diagrams show data points 32 in a schematic environment space 34. The data points 32 are characterized by a set of elements each corresponding to a dimension of the integration space 34. The integration space 34 is exemplarily represented by a three-dimensional space although it will be clear to those skilled in the art that the integration space 34 may have more dimensions. The integration space 34 is divided into two classes by a boundary 36 which separates the “non-emergency” class (circles) and the “emergency” class (crosses).

[0063] [Fig.3] shows a random forest approach. Random forest is a system known for its robustness, the ability to process large data, resistance to overfitting and relatively low sensitivity to hyperparameters.

[0064] A labeled data set is used for training and validation. The labeled data set comprises data points 32 that are obtained by letting different test subjects interact with a simulated environment. The test subjects are different fixed driving situations or driving tasks. In particular, the different driving situations and driving tasks can be simulated based on contextual information data 26. For each configuration of contextual information data 26, the vehicle dynamic data 24 generated by the driving test subject is recorded as data point 32 during the simulation. Then during the simulation, a label is manually set to the emergency level. The data points 32 can be recorded in real time or at a fixed sampling rate, typically 10 seconds, 30 seconds, or one minute.

[0065] From the labeled dataset, each data point 32 is associated with a class label such as "non-emergency" (circle) and "emergency" (cross). Random forest introduces randomness in two ways.

[0066] First, a subset of the dataset is randomly selected (with replacement) to create multiple training datasets. This process is called "extraping" or "bagging." Each of the subsets is referred to as a "bootstrap example."

[0067] For each bootstrap example, a decision tree is constructed. Decision trees are used for binary classification, i.e., two classes, non-emergency and emergency. During training, each tree learns a mapping between input features based on vehicle dynamic data 24, contextual information data 26 and class labels.

[0068] At each node of the tree, a random subset of features is considered for separation. This introduces another layer of randomness and helps avoid overfitting. The tree is grown to a stopping criterion such as a maximum depth or a minimum number of samples per leaf.

[0069] After constructing multiple decision trees, the next step is to aggregate their prediction. In the case of classification, each decision tree votes for the class it predicts. For binary classification, the class with the majority of votes becomes the predicted class for the input data point 32.

[0070] Random forest includes a built-in evaluation mechanism called "out-of-bag" (OOB) error. Since bootstrap sample is created by random sampling, some 32 data points are left "out of the bag" during the construction of each tree. The OOB error is calculated by evaluating each 32 data point with only the trees for which it was an out-of-bag sample. This gives the estimate of the model performance without requiring separate validation sets.

[0071] Random forest provides a measure of feature importance. It calculates the average decrease in impurity (usually Gini impurity) produced by each feature across all decision trees. Features that consistently yield better separations are considered the most important.

[0072] [Fig.4] shows a kernel density estimation approach. KDE-type binary classification allows capturing complex, non-binary relationships between data. It does not rely on a predefined decision boundary making it more flexible for manipulating data with nested patterns. KDE classification is particularly useful if one has prior knowledge of the data distribution or if the data distribution cannot be easily separated by a linear boundary.

[0073] Binary classification with KDE kernel density estimation is an approach to classify 32 data points into emergency and non-emergency based on their estimated probability density. KDE classification uses the underlying data distribution to make predictions.

[0074] We again start from a labeled dataset containing non-emergency and emergency data points, each data point 32 having a set of features. For training, the dataset is separated into a training set and a test set to evaluate the performance of the classifier.

[0075] For each class, an independent estimate of the probability density function (PDF) is made using KDE which consists of selecting a kernel function (e.g., a Gaussian or Epanechnikov function), choosing an appropriate bandwidth parameter for the width of the kernels and calculating the KDE for each class with the respective training data. The bandwidth parameter is chosen using techniques such as cross-validation.

[0076] In operation, the KDE estimation is fed with features based on the vehicle dynamic data 24 and the contextual information data 26; for each data point 32, the probability density for the data point 32 is calculated each time in two specific KDE classes. This results in a density value for each class.

[0077] To classify the measured data point 32, it is compared to two density values. The point is assigned to the class with the highest estimated probability density. In other words, if the estimated non-emergency density is the highest, the data point 32 is classified as a "non-emergency" class. Otherwise, it is classified as an "urgent" class.

[0078] To evaluate the performance of the KDE classifier, it is applied to the test dataset and the predictions are compared to the true class labels. Preferably, standard classification metrics such as precision, recall, Fl level and ROC curve are used to evaluate the performance of the KDE classifier.

[0079] [Fig. 5] shows a method for determining the driver's urgency level. The method is preferably carried out by the driver assistance system 18.

[0080] In the preparatory step SI, a trained classifier is loaded, generally a random forest classifier, or the KDE classifier. In addition, an abnormality counter "a" initialized to a = 0 and a time counter "k" initialized to k = 1 are initialized.

[0081] In the starting step S2, the driving assistance system 18 is launched.

[0082] In a data collection step S3, the driving assistance system 18 collects the sensor data 20 and the vehicle dynamic data. If applicable, the driver assistance system 18 also takes the contextual information data 26 via the telecommunications link 22 from other vehicles 16 and / or the internet. The contextual information data 26 includes the vehicle dynamic data but for other vehicles 16 and / or additional information such as traffic data and road condition data. The vehicle dynamic data and the contextual information data 26 are collected globally every minute.

[0083] In the data collection step S3, characteristics of the vehicle dynamic data 24 and the contextual information data 26 are extracted. possible to construct a feature vector by concatenating the vehicle dynamic data 24 and the contextual information data 26. The feature vector is preferably based on the speed, acceleration and / or jerk, the average speed Vavg of the own vehicle 10, the average jerk Javg of the own vehicle 10, the speed deviation AVavg of the average speed Vavg of the own vehicle 10 from the average boundary speed Vavg_road and the jerk deviation AJavg of the average jerk Javg of the own vehicle 10 from the average road jerk Javg_road.

[0084] The following characteristics may also be suitable for the described application: maximum speed, maximum speed deviation AVavg, vehicle jerk range, maximum acceleration, maximum jerk, geometric mean of speed deviation AVavg, maximum acceleration deviation, third quartile of speed deviation AVavg, maximum jerk deviation AJavg, standard deviation of acceleration, standard deviation of speed, third quartile of acceleration deviation, average acceleration, speed deviation range AVavg and acceleration deviation range.

[0085] In the classification step S4, the feature vector is classified with the trained classifier. If the classifier finds an anomaly, the anomaly counter “a” is incremented by one unit; otherwise the anomaly counter “a” remains unchanged.

[0086] In the level step S5, an emergency level “z” is determined by dividing the current value of the anomaly counter “a” by the current time step “k” or in the form z = p. Then the time step “k” is increased by one unit.

[0087] In a comparison step S6, the current emergency level “z” is compared to the distribution of the historical emergency level Ui(Z) representing the driver 12. If a predetermined threshold is exceeded by the current emergency level “z”, the driving assistance system 18 notes that the driver 12 is in an emergency.

[0088] In a control step S7, a control signal is generated based on the result of the comparison step S6. If the driver 12 is considered not to be in an emergency, the control signal does not change the behavior of the driver assistance system 18 or of the own vehicle 10. If the driver 12 is considered to be in an emergency, the control signal is generated so that the driver assistance system 18 informs the driver 12 of his driving style by displaying a warning. If this is also possible, the control signal limits the driver assistance system 18 to the maximum speed and / or acceleration of the own vehicle 10 to increase the sensitivity of the collision detection and / or to increase the braking power.

[0089] In a terminal step S8, the driving assistance system 18 looks at the data collection step S3 unless, for example, the driver 12 stops his own vehicle 10.

[0090] NOMENCLATURE OF MAIN ELEMENTS

[0091] 10 Own vehicle

[0092] 12 Driver

[0093] 14 Route

[0094] 16 Other vehicle

[0095] 18 Driving assistance system

[0096] 20 Sensor

[0097] 22 Telecommunications link

[0098] 24 Vehicle dynamic data

[0099] 26 Contextual information data

[0100] 28 Driver profile

[0101] 30 Historical Emergency Data

[0102] 32 Data Point

[0103] 34 Integration space

[0104] 36 Border

Claims

Claims

1. A computer-implemented method for determining an emergency level of a driver (12) in his own vehicle (10) based on contextual information data (26), the method comprising: a) inputting vehicle dynamic data (24) representing at least one quantity related to the movement of the own vehicle (10), b) inputting the contextual information data (26) representing at least one property external to the own vehicle (10), c) generating a feature vector of his vehicle dynamic data (24) and contextual information data (26), d) classifying this feature vector into an emergency class or a non-emergency class and tracking the quantity of occurrences of the emergency class with an anomaly counter, and e) determining an emergency level based on the value of the anomaly counter.

2. The method of claim 1, wherein in state a), the vehicle dynamic data (24) comprises any of the following data: speed of the own vehicle (10), acceleration of the own vehicle (10) and jerks of the own vehicle (10).

3. A method according to any one of the preceding claims, wherein in step b), at least one property selected from any of the data: weather data, traffic jam data, road condition data, traffic data and vehicle dynamic data for at least one other vehicle (16).

4. A method according to any preceding claim, wherein in step c), the feature vector is generated to comprise any of the data: maximum speed, maximum speed deviation AVavg, vehicle jerk range, maximum acceleration, maximum jerk, geometric mean of speed deviation AVavg, maximum acceleration deviation, third quartile of speed deviation AVavg, maximum jerk deviation AJavg, standard deviation of acceleration, standard deviation of speed, third quartile of acceleration deviation, average acceleration, speed deviation range AVavg and acceleration deviation range and at least one statistical characteristic extracted from the vehicle dynamic data of the own vehicle and / or at least one other vehicle.

5. A method according to any one of the preceding claims, wherein in step d), the classification is carried out according to the binary supervision method or the multi-class classification method, wherein the binary classification method and the multi-class classification method output a result indicating a normal state or a result of an emergency, - the multi-class classification method further providing outputs of at least one result indicating another class of behavior of the driver (12).

6. The method of claim 5, wherein the classification method comprises the random forest approach or the XGBoost approach.

7. Method according to any one of claims 1 to 4, according to which in step d), the classification is made by the unsupervised classification method.

8. The method of claim 7, wherein the classification method comprises a kernel density estimation method for classifying the feature vector into an emergency class or a non-emergency class.

9. A method according to any preceding claim, wherein in step d), the urgency level is determined by moving the average approach which relates to dividing the anomaly counter by a current time step, measured from the start of a journey with one's own vehicle (10).

10. A method according to any preceding claim, wherein step e) comprises comparing the urgency level to historical urgency level data, and determining that the driver (12) is in an emergency, the emergency level exceeding a predetermined threshold, preferably provided by the historical emergency level.

11. A method applied by a computer for controlling a driver assistance system (18) for its own vehicle (10), the method comprising: - applying a method according to any one of the preceding claims to determine an emergency level, comparing the emergency level with historical emergency level data and determining that the driver (12) is in an emergency, the emergency level exceeding a predetermined threshold provided by the historical emergency level, - generating a control signal for the own vehicle (10) to display a warning message to increase the braking power, to increase the braking force, to increase the sensitivity of the collision sensors and / or to decrease the maximum speed and / or acceleration of the own vehicle (10).

12. Driver assistance system (18) for his own vehicle (10) comprising means for applying the method according to claim 11.

13. Own vehicle (10) comprising a driving assistance system (18) according to claim 12.

14. A computer program comprising instructions which when the program is executed by a computer, cause the computer to apply the method according to any one of claims 1 to 10 or when the method is executed by a driver assistance system (18) according to claim 12, cause the driver assistance system (18) to apply the method according to claim 11.

15. A computer-readable medium or data-carrying signal comprising the computer program of claim 14.

Images