Method and system for operating a lane change assistant in a vehicle

Abstract

Method for operating a lane change assistant (102) of a vehicle (104), in which a) Sensor data from sensors (110) of vehicles (106) of a vehicle fleet (108) are recorded, wherein at least part of the sensor data is recorded as a time series of data points; b) based on the sensor data, a large number of lane change data are generated, each of which is assigned to a lane change of one of the vehicles (106) of the vehicle fleet (108), wherein, from the part of the sensor data which has been recorded as a time series of data points, at least those data points which have been recorded at the beginning of the lane change and at the end of the lane change are used to generate the lane change data; c) Fleet data is generated based on lane change data; d) the fleet data are processed using cluster analysis to classify the lane changes into at least two groups (500, 502, 504, 506, 508) that differ in their lane change scenario; and e) is operated on the basis of cluster analysis of the lane change assistant (102).

Description

[0001] The invention relates to a method for operating a lane change assistant in a vehicle. The invention further relates to a system for operating a lane change assistant in a vehicle.

[0002] Lane change assistants are driver assistance systems that enable safer, automatic lane changes on highway-like roads. Besides increased driving comfort, one of the main reasons for the development of lane change assistants is that traffic jams and accidents are often caused by inappropriate or excessively frequent lane changes. However, to improve lane change assistants, it is essential to be able to accurately characterize various lane change scenarios in order to determine, for example, whether a lane change is appropriate or not.

[0003] Several methods for classifying vehicle lane changes are known from the prior art. These methods typically rely on the experience of experts or the perception of drivers to perform a rule-based or model-based classification. When sensor data is used as the basis for classification, it often originates from a very small group of vehicles, typically comprising a few dozen, rarely up to a few thousand, or is obtained through simulations in driving simulators. The predictive power of the known methods is correspondingly limited.

[0004] A method for identifying and classifying lane changes performed automatically by a vehicle is disclosed in CN 1 14 550 121 A. This method employs an agglomerative hierarchical clustering technique.

[0005] CN 1 14 169 371 A discloses a method for classifying driving styles using clustering algorithms. Data from a driving simulator are used as the basis.

[0006] CN 1 11 461 185 A further discloses a method for analyzing driver behavior using a k-means algorithm. Data from a GPS system is used as the basis.

[0007] The publication DE 10 2020 215 780 A1 describes a method for selecting an at least partially automated driving process using a vehicle's driver assistance system.

[0008] Publication WO 2019 / 122 951 A1 concerns a lane planning system for autonomous vehicles. This system receives sensor data capturing road surface images of the road on which the autonomous vehicle is located. Based on this sensor data, the system identifies the current lane occupied by the autonomous vehicle. Lane control for the autonomous vehicle is planned according to a lane control model, based on the identified current lane and self-awareness parameters. The self-awareness parameters are used to predict the operational capability of the autonomous vehicle with respect to its current location. The lane control model is generated based on recorded human driving data to achieve human-like lane control behavior in various scenarios.

[0009] The German patent application DE 10 2023 113 922 A1 relates to a driver assistance system for a vehicle, comprising a detection module configured to detect a lane change initiated by a driver; a classification module configured to classify the detected lane change as either an intentional or unintentional lane change, wherein the classification module implements a trained classification algorithm for the classification; and at least one control module configured to control at least one vehicle function based on the classification of the lane change.

[0010] The object of the invention is to provide a method and system for operating a lane change assistant of a vehicle which enables an improved classification of lane change scenarios compared to the known prior art.

[0011] This problem is solved by a method having the features of claim 1 and by a device having the features of the independent device claim. Advantageous embodiments are specified in the dependent claims.

[0012] The proposed method for operating a vehicle's lane change assistant involves at least the following steps: a) Sensor data from sensors of vehicles in a fleet are acquired. At least some of the sensor data is acquired as a time series of data points. b) Based on the sensor data, a large number of lane change data points are generated, each corresponding to a lane change performed by one of the vehicles in the fleet. From the portion of the sensor data acquired as a time series of data points, at least those data points acquired at the beginning and end of the lane change are used to generate the lane change data. c) Fleet data is generated based on the lane change data.d) The fleet data is processed using cluster analysis to divide the lane changes into at least two groups that differ in their lane change scenario. e) The lane change assistant is operated based on the cluster analysis.

[0013] The proposed method is based on sensor data acquired during numerous lane changes by the vehicles in the fleet. This sensor data is then preprocessed by generating lane change data. To reduce the amount of data to be processed, the data points from the time series of sensor data acquired are those recorded at the beginning and end of the lane change. In other words, for each time series and lane change, at least the two data points recorded at the beginning and end of the lane change are processed. This selection is based on the inventive finding that the time points at the beginning and end of a lane change are particularly suitable for characterizing the lane change. Additionally, data points before, after, or during the lane change can also be used.For example, additional data points can be used that have been recorded a predetermined time before the start of the lane change and / or a predetermined time after the end of the lane change, for example 2, 3 or 5 seconds each, so that no more than four data points are used per time series and lane change.

[0014] Fleet data relating to the vehicle fleet is then generated from the lane change data associated with each individual lane change. This fleet data can be generated, for example, by simply aggregating the lane change data. Preferably, however, a selection is made of the lane changes whose lane change data is to be aggregated into the fleet data in order to further reduce the amount of data to be processed. For example, only lane changes performed by a specific vehicle type, model, or model series can be considered. This could include, for instance, the vehicle type, model, or model series in which the lane change assistant is used. Further examples of such a selection are described below using exemplary embodiments of the method.

[0015] The fleet data is then processed using cluster analysis. This analysis divides the lane changes described by the fleet data into at least two groups (clusters) that differ in their lane change scenario, so that each recorded lane change is assigned to a group. The different clusters, in turn, allow for the characterization of various lane change scenarios. For example, different lane change scenarios can differ in their duration, the number and position of other vehicles in the vicinity, and / or the longitudinal and lateral acceleration during the lane change. Since the fleet data is the result of preprocessing that reduces the amount of data compared to the recorded sensor data, the proposed method can process a large number of lane changes from the vehicle fleet to perform the classification.For example, around 100,000 lane changes are processed in the form of lane change data. A large database for cluster analysis, in turn, makes the classification particularly meaningful. The method thus enables a robust and accurate classification of lane change scenarios, on which the lane change assistant is based.

[0016] In one embodiment, at least one driving scenario of the vehicle is classified based on cluster analysis. The lane change assistant is then operated based on this classification. In this embodiment, for example, the vehicle's current driving situation is captured in the form of sensor data. The driving scenario is generated from this sensor data, for example by aggregating the sensor data, and then classified. The classification, i.e., the result of the classification, can then be used by the lane change assistant to make robust and reliable decisions.

[0017] In another embodiment, the classification determines whether the lane change assistant is activated. In this embodiment, the classified driving scenario is used to determine, for example, whether the vehicle is currently performing a dangerous lane change. If a dangerous lane change is detected, the lane change assistant is activated to support the driver. This makes lane changes even safer by using the classification.

[0018] In another embodiment, the classification determines whether a lane change should occur. The lane change assistant operates based on this decision. Certain sensor data values ​​and derived parameters reliably indicate an imminent lane change, such as the activation of a turn signal. Based on such signals, a reliable decision can be made to initiate a lane change, and the lane change assistant can be automatically activated. The specific values ​​indicating an impending lane change vary from lane change scenario to lane change scenario. These specific values ​​are determined for each group using cluster analysis.

[0019] In another embodiment, the system uses classification to determine whether a lane change is appropriate. The lane change assistant then operates based on this decision. A lane change might be inappropriate, for example, because it is performed too quickly and / or in heavy traffic. This information can be reliably derived from the driving scenario, which in turn is determined based on sensor data and the parameters derived from it. If it is determined that a lane change is inappropriate, the lane change assistant operates accordingly. For example, the lane change assistant is activated to support the driver, or the lane change is prevented by the lane change assistant. The criteria for when a lane change is inappropriate vary from one lane change scenario to another.The specific values ​​of sensor data that indicate an inappropriate lane change in a lane change scenario are determined based on cluster analysis for the different groups.

[0020] In another embodiment, a machine learning algorithm is trained based on the results of the cluster analysis to generate a classifier. This classifier is trained to classify vehicle driving scenarios based on input data corresponding to the vehicle's sensor data and / or derived variables. The lane change assistant operates using this classifier. Classification with the machine learning algorithm is particularly efficient because machine learning algorithms are capable of reliably identifying complex patterns even in large datasets. The use of the classifier therefore makes the lane change assistant exceptionally powerful.

[0021] The machine learning algorithm is trained using a training dataset. This dataset comprises a variety of training data and is preferably generated based on a first subset of the recorded lane changes. Each training dataset corresponds to a lane change from the first subset and includes the sensor data and / or derived quantities related to that lane change. Preferably, the same type of sensor data and / or derived quantities are used for training as are used as input for the classifier. The training dataset also includes, as a label, the group into which the respective lane change was assigned as a result of cluster analysis. Similarly, a test dataset can be generated based on a second subset of the recorded lane changes. Using the test dataset as input, the training of the machine learning algorithm can be verified.The first part and the second part are preferably disjoint, i.e., they have an empty intersection.

[0022] The preferred machine learning method is a random forest or another tree-based machine learning approach. Tree-based methods are intuitive and easy to interpret, making it very simple to verify the classification. Alternatively, regression, particularly linear regression, can be performed. Regressions are especially easy to implement and can be performed with minimal resource consumption. They can therefore be easily carried out using the vehicle's existing resources. Furthermore, a neural network can also be used as the machine learning method. Neural networks are capable of recognizing and modeling complex and nonlinear patterns in large datasets. Neural networks are therefore particularly suitable when a large number of lane changes are available as a training basis.

[0023] In another embodiment, the input data comprises data from at least one of the following vehicle sensors: a lateral acceleration sensor, a longitudinal acceleration sensor, a distance sensor, an external camera, an internal camera, and a position sensor. The sensor data used to generate the lane-change data also preferably includes data from at least one of the aforementioned sensors. The lateral acceleration sensor and the longitudinal acceleration sensor can be used to detect the vehicle's lateral and longitudinal acceleration, respectively. The distance sensor measures the vehicle's distance to other vehicles. For example, the distance to a vehicle ahead can be determined using the distance sensor. The external camera can capture images of the area around the vehicle.Based on these images and a suitable image recognition method, such as a machine learning algorithm, a wealth of information can be captured, including current weather conditions, visibility, and daylight levels. Furthermore, the system can determine, for example, the number and type of objects in the vehicle's vicinity, the type of lane markings, the number of lanes, and whether the vehicle is entering or exiting a construction zone. The interior camera can capture images of the vehicle's interior and its occupants, particularly the driver. These images can be used to determine, for instance, the driver's level of attention and whether their hands are on the steering wheel.The position sensor is designed to determine the vehicle's geographic position and includes, for example, a receiver for a global navigation satellite system (GNSS) and / or other sensors such as gyroscopes. Based on the geographic position, the geographic region in which the vehicle is located can be determined. Furthermore, the geographic position can be compared with map data to determine, for example, the number of lanes and the type of road surface at the vehicle's location. Based on the aforementioned information, which is obtained using the vehicle's sensors, the current driving scenario can be described and classified very precisely.

[0024] In another embodiment, the input data includes the state of at least one of the following vehicle systems: a turn signal indicator and a driver assistance system, in particular the lane change assist. The sensor data used to generate the lane change data also preferably includes the state of at least one of the aforementioned vehicle systems. Based on this information, it can be determined, for example, whether a lane change should be initiated by activating the turn signal indicator. Furthermore, it can be determined whether a driver assistance system, such as the lane change assist, is currently actively supporting the driver. This makes it possible to decide, for example, whether the lane change assist needs to be activated or not.

[0025] In another embodiment, the input data comprises vehicle data that corresponds to at least one piece of vehicle-related information. For example, the vehicle data corresponds to at least one of the following pieces of vehicle information: longitudinal acceleration, lateral acceleration, vehicle speed, the state of a turn signal indicator, and the state of a driver assistance system. The sensor data used to generate the lane change data also preferably includes at least one of the aforementioned pieces of vehicle information. In particular, the lateral acceleration can be used to determine the point in time at which the lane change begins. The longitudinal acceleration and / or vehicle speed can be used to determine whether the vehicle is accelerating in the direction of travel. Based on this information, for example, an impending overtaking maneuver can be identified.The aforementioned vehicle data and derived parameters characterize the current driving scenario of the vehicle and thus enable a robust classification.

[0026] In another embodiment, the input data each comprise environmental data corresponding to at least one piece of environmental information related to the vehicle's surroundings. For example, the environmental data corresponds to at least one of the following pieces of environmental information: the distance of the vehicle performing the lane change to other vehicles, the number of objects in the vicinity of the vehicle performing the lane change, the position of objects in the vicinity of the vehicle performing the lane change, the type of objects in the vicinity of the vehicle performing the lane change, the number of lanes, the distance of the vehicle performing the lane change to a lane marking, the type of lane marking, the road type, the current weather conditions, the current visibility, and the current daylight conditions.The sensor data used to generate the lane change data preferably also includes at least one of the aforementioned environmental information.

[0027] Preferably, the values ​​of the aforementioned environmental information are continuously determined. For example, the number of lanes can be continuously determined. This allows the system to ascertain whether the roadway widens, for example, after roadworks, or narrows, for example, at the beginning of roadworks. In another example, the position of vehicles relative to the vehicle performing the lane change is continuously determined. This allows the lane change assistant to essentially track the other vehicles. From this, it can be determined, for example, whether a lane change should be performed as part of an overtaking maneuver. The aforementioned environmental data and the quantities derived from it characterize the vehicle's current driving scenario and thus enable robust classification.

[0028] In another embodiment, the input data includes driver data that corresponds to at least one piece of driver information related to the vehicle's operator. The sensor data used to generate the lane change data also preferably includes at least one piece of driver information. For example, the driver's gaze direction or whether the driver has their hands on the steering wheel can be determined. Furthermore, it can be determined whether the driver is attentive. The aforementioned driver data and derived parameters are helpful in characterizing the vehicle's current driving scenario and thus enable an even more robust classification.

[0029] In another embodiment, cluster analysis is performed as unsupervised learning. Preferably, a k-means algorithm is used to carry out the cluster analysis. However, other classification algorithms can also be used for cluster analysis, for example, other partitioning clustering methods such as affinity propagation, hierarchical clustering methods, density-based clustering methods such as DBSCAN or OPTICS, or combined methods such as Bayesian-Gaussian mixture models, spectral clustering, or BIRCH. All these methods allow the fleet data to be reliably classified in order to identify the different groups of lane-change scenarios.

[0030] In another embodiment, sensor data is continuously acquired, and lane change and fleet data are regularly regenerated and / or updated. In particular, cluster analysis is also performed regularly based on the newly generated and / or updated fleet data. This continuously expands and updates the dataset used for classifying lane change scenarios, enabling, for example, a better representation of emerging trends. Based on this continuously updated classification, the lane change assistant can operate more reliably.

[0031] In another embodiment, the sensor data and / or lane change data are anonymized before the fleet data is generated. This makes it possible, in compliance with data protection regulations, to record a large number of lane changes and thus obtain a more robust classification, on the basis of which the lane change assistant can be operated more reliably.

[0032] The invention further relates to a system for operating a vehicle's lane change assistant. The system comprises a receiver unit configured to receive sensor data from sensors of vehicles in a fleet. The system also comprises a processing unit configured to generate a plurality of lane change data based on the sensor data, each data point corresponding to a lane change performed by one of the vehicles in the fleet. The processing unit is further configured to generate fleet data based on the lane change data and to process the fleet data using cluster analysis to classify the lane changes into at least two groups that differ with respect to their lane change scenario. The system further comprises at least one vehicle's lane change assistant configured to be operated based on the results of the cluster analysis.

[0033] The receiving unit and the processing unit can be implemented in various ways. For example, the receiving unit and the processing unit can individually or jointly comprise one or more servers maintained by the system operator. Cloud computing, especially serverless computing, offers a particularly advantageous implementation option. These methods allow processing to take place in a cloud without the operator having to worry about server maintenance or scaling. A key advantage here is cost efficiency, as billing is based on actual computing time used.

[0034] The system has the same advantages as the claimed method. In particular, the system can be further developed with the features described in this document in connection with the method. Likewise, the method described above can be further developed with the features described in this document in connection with the system.

[0035] Exemplary embodiments of the invention are explained in more detail below with reference to the figures. These show: Fig. 1 a schematic representation of a system for operating a lane change assistant of a vehicle according to one embodiment; Fig. 2 a flowchart of a procedure for operating a lane change assistant of a vehicle according to an embodiment; Fig. Figure 3 shows a flowchart of an exemplary sub-procedure of the procedure according to Fig. 2, in which the vehicle's lane change assistant is operated based on the result of a cluster analysis; Fig. 4 a grid-like pattern of 15 discrete distance ranges which can be used to divide distances detected by vehicle distance sensors into discrete distance ranges; and Fig. 5 a schematic representation of five exemplary groups resulting from the procedure according to Fig. 2 were identified.

[0036] Fig. Figure 1 shows a schematic representation of a system 100 for operating a lane change assistant 102 of a vehicle 104 according to one embodiment. The system 100 is configured to first classify a large number of lane changes performed by vehicles 106 of a vehicle fleet 108 into two or more groups that differ with respect to their lane change scenario. The system 100 uses this classification as the basis for operating the lane change assistant 102. Five exemplary groups 500, 502, 504, 506, 508 of different lane change scenarios are shown with reference to Fig. 5 below described in more detail.

[0037] The vehicle fleet 108 comprises a large number of vehicles 106, for example, the vehicles 106 of a specific manufacturer in a geographical region, such as a particular country. The vehicles 106 perform a large number of lane changes over a specific period, such as a year or several months. Before, during, and after the lane changes, sensors 110 of the vehicles 106 record sensor data. For example, lateral acceleration sensors record the lateral acceleration of the vehicles 106, longitudinal acceleration sensors record the longitudinal acceleration of the vehicles 106, speedometers record the vehicle speed, and external cameras capture images of the surroundings of the respective vehicle 106. At least one type of sensor data, such as lateral and longitudinal accelerations as well as vehicle speed, is recorded as a time series of data points, each assigned to a specific point in time.Additionally, the states of vehicle systems 112 of vehicles 106 of the vehicle fleet 108 can also be recorded as sensor data, for example, the state of a turn signal indicator and a lane change assistant. The states of vehicle systems 112 can also be recorded as a time series. The sensor data is uniquely assigned to each vehicle 106 of the vehicle fleet 108.

[0038] The sensor data recorded by the vehicles 106 of the vehicle fleet 108 are transmitted to the system 100. For example, each vehicle 104 of the vehicle fleet 108 transmits the sensor data it has recorded to the system 100 at regular intervals. The transmission takes place, for example, via a remote data transmission network 114, such as the internet and / or a mobile network. A receiving unit 116 of the system 100 is configured to receive the sensor data. The actual processing of the sensor data is then carried out by a processing unit 118 of the system 100. Fig. In Figure 1, the receiving unit 116 and the processing unit 118 are purely exemplary elements of a common computing unit, such as a server or a cloud service. However, the receiving unit 116 and the processing unit 118 can also be configured as independent computing units, for example, as different servers or cloud services.

[0039] The processing unit 118 is designed to first perform preprocessing of the sensor data. This reduces the amount of sensor data to decrease the volume of data to be processed. Specifically, the processing unit 118 is designed to further process only those data points from the sensor data, which has been recorded as a time series, that correspond to predetermined points in time, such as the beginning and end of a lane change. However, the processing unit 118 can also process additional points in time before, during, or after a lane change. For example, the processing unit 118 processes those data points corresponding to the times 2 seconds before a lane change, at the beginning of the lane change, at the end of the lane change, and 2 seconds after the lane change.On the other hand, the processing unit 118 can be configured to determine derived quantities from the raw sensor data, which can then be used as the basis for further processing instead of the raw sensor data itself. For example, the distances of a vehicle 106 performing a lane change to other vehicles 106 or objects can be divided into discrete distance ranges to reduce the number of possible values ​​and facilitate further processing. Exemplary distance ranges are shown below. Fig. As described in section 4. Furthermore, for example, a random selection of lane changes can be made for further processing to reduce the amount of data. As a result of the preprocessing, processing unit 118 generates a large number of lane change data, each uniquely assigned to a lane change. From the lane change data, processing unit 118 generates fleet data, for example by aggregating the individual lane change data.

[0040] The processing unit 118 is further configured to process the fleet data using cluster analysis in order to classify the lane changes performed by the vehicles 106 of the vehicle fleet 108 into two or more groups 500, 502, 504, 506, 508. The result of the cluster analysis is a classification of the lane changes performed by the vehicles 106 of the vehicle fleet 108, which is used by the system 100 to operate the lane change assistant 102 of the vehicle 104. For example, at least one driving scenario is recorded using sensors 116 of the vehicle 104 and, if applicable, other vehicle systems 118, and classified using the classification in order to operate the lane change assistant 102. The vehicle fleet 108 and the system 100 thus carry out a procedure for operating a lane change assistant 102 of a vehicle 104. This procedure is described below with reference to Fig. 2 described in more detail.

[0041] Fig. Figure 2 shows a flowchart of the procedure for operating a lane change assistant 102 of a vehicle 104 according to one embodiment. The procedure can be implemented, for example, using the method described above. Fig. The system described in section 1 will be carried out in 100.

[0042] The process is initiated in step S200. In step S202, sensor data is collected by sensors 110 of vehicles 106 in the vehicle fleet 108. At least some of the sensor data is recorded as a time series of data points. Step S202 is preferably executed for a predetermined period, for example, a calendar year. The sensor data can be collected anonymously, in particular, to enable the processing of large data volumes in compliance with data protection regulations. Step S202 can be repeated regularly to continuously update and maintain the data basis for the process.

[0043] In step S204, the sensor data is preprocessed to generate the multitude of lane change data. This preprocessing includes, at a minimum, reducing the sensor data, which was acquired as a time series, to those data points that correspond to the predetermined time points. Specifically, this means that at least the time points at the beginning and end of a lane change are considered in the further course of the process, as these are particularly informative. Preferably, in the method according to the invention, those data points are used to generate the lane change data that correspond to a predetermined time before a lane change, at the beginning of the lane change, at the end of the lane change, and a predetermined time after the lane change. The predetermined time is, for example, 2, 3, or 5 seconds. By reducing the number of data points, the amount of data to be processed can be significantly reduced.Preprocessing can further include a selection of lane changes that serve as the basis for subsequent processing, thereby reducing the amount of data to be processed. The selection of lane changes can be random. For example, individual lane changes can be randomly selected from a large number of lane changes based on unique identification numbers assigned to them. Alternatively, only those lane changes that meet a predetermined criterion can be selected for further processing. For example, only lane changes performed on selected days, in a specific geographic region, and / or by a specific vehicle model are considered. A further random selection can then be made from the lane changes that meet the predetermined criterion.Conversely, a random selection of lane changes can also be restricted to those that meet the predetermined criterion. As part of step S204, the sensor data can also be anonymized, provided this has not already been done in step S202.

[0044] The preprocessing in step S204 can further include generating quantities derived from the raw sensor data. For example, distances detected by the vehicles' distance sensors can be divided into discrete distance ranges. The detected distances can, for example, be divided into a grid of 400 discrete distance ranges, which is then further subdivided into 15 discrete distance ranges. Fig. Figure 4 shows that the discrete distance ranges can be used to characterize the position of vehicles and other objects relative to the vehicle 106 performing the lane change. Furthermore, the number of lanes on the section of road where the lane change is performed can be determined. This can be done, for example, by evaluating images from an external camera and / or by comparing position data acquired by the vehicle 106's position sensors with map data. Similarly, it can also be determined in which lane a vehicle 106 is located at the beginning and end of a lane change.

[0045] The lane change data corresponds to a set of information that can be divided into three categories. The first category comprises vehicle information related to the vehicle 106 performing the lane change, such as the current speed as determined by a speedometer and the state of a turn signal before the lane change begins. The second category comprises environmental information related to the surroundings of the vehicle 106 performing the lane change, such as the type and number of objects in the vicinity of the vehicle 106 performing the lane change, as well as the weather conditions at the time the lane change is performed.A third category includes driver information relating to the driver of vehicle 106 performing the lane change, such as whether the driver has his hands on the steering wheel or not, which can be determined using images from an interior camera and a suitable image recognition algorithm.

[0046] In step S206, fleet data is generated from the lane change data relating to a single lane change. If a selection of lane changes to be considered was made in step S204, the fleet data is generated by aggregating the lane change data corresponding to those lane changes. If, however, no selection of lane changes to be considered was made in step S204, all lane change data is aggregated to generate the fleet data.

[0047] In step S208, the fleet data are processed using cluster analysis to divide the lane changes under consideration into at least two groups 500, 502, 504, 506, 508 (clusters) that differ in their lane-change scenario. The cluster analysis is preferably performed using a k-means algorithm as unsupervised learning. To determine the optimal number of groups 500, 502, 504, 506, 508 (clusters) for the k-means algorithm, a number of well-known techniques can be used, such as an iVAT algorithm, a silhouette coefficient, or the elbow method. With reference to Fig. Figure 4 describes an exemplary result of the cluster analysis for a data set of 104,000 lane changes.

[0048] In the optional step S210, a machine learning algorithm is trained based on the results of the cluster analysis to create a classifier that is trained to classify driving scenarios of vehicle 104 based on input data corresponding to sensor data of vehicle 104 and / or derived quantities. For example, a Random Forest or other tree-based models can be trained to classify lane-change scenarios.

[0049] To train the machine learning algorithm, a training dataset is first generated, comprising a variety of training data. This training data is generated, for example, based on an initial portion of the lane changes recorded in step S202. The training data includes the sensor data related to each lane change, recorded in step S202, and / or the derived variables generated in step S204. It is not necessary to use all types of sensor data or all derived variables. It is sufficient to use only those types of sensor data or derived variables that will later be used as input data by the machine learning algorithm. The training data also includes the group 500, 502, 504, 506, 508, into which the respective lane change was categorized as a result of the cluster analysis, as its label.Based on a second part of the recorded lane changes, a test data set can be generated which can be used to verify the machine learning process.

[0050] In step S212, the lane change assistant 102 of vehicle 104 is operated based on the result of the cluster analysis determined in step S208. If step S210 was performed, the lane change assistant 102 is operated using the classifier. An exemplary sub-process that can be executed in step S212 is described below with reference to Fig. 3 described. The procedure is then terminated in step S214.

[0051] Fig. Figure 3 shows a flowchart of an exemplary sub-procedure of the procedure according to Fig. 2, in which the lane change assistant 102 of the vehicle 104 is operated based on the result of the cluster analysis.

[0052] The sub-procedure is started in step S300. In step S302, a driving scenario for vehicle 104 is determined. For this purpose, data from sensors 116 of vehicle 104 can be collected. For example, data from a lateral acceleration sensor, a longitudinal acceleration sensor, a distance sensor, an external camera, an internal camera, and / or a position sensor of vehicle 104 are collected. Derived quantities can also be generated from the raw data of the sensors 116. For example, the images from the external camera can be processed using a suitable image processing algorithm to detect objects in the vicinity of vehicle 104. Furthermore, detected distances can be divided, for example, into the grid-like raster 400 of 15 discrete distance ranges, which is shown in Fig. Figure 4 shows that the states of vehicle systems 118 of the vehicle 104, for example the turn signal, can be recorded as data. The collected data and the quantities derived from them can then be aggregated to form the driving scenario. This is similar to the lane change data recorded in step S204 of the procedure according to... Fig. The driving scenario generated in step 2 can encompass a wide range of information. Specifically, it includes vehicle information related to vehicle 104, environmental information related to the surroundings of vehicle 104, and / or driver information related to the driver of vehicle 104. In step S304, the previously recorded driving scenario is then assigned to one of the groups 500, 502, 504, 506, or 508 determined in step S208, i.e., classified. The classifier generated in step S210 can be used for this purpose.

[0053] Finally, in step S306, the Lane Change Assist 102 is activated based on the classification of the driving scenario performed in step S304. For example, the driving scenario can first determine whether a lane change has been initiated. In another example, it is determined whether the driving scenario requires the Lane Change Assist 102 to be activated to support the driver because a particularly dangerous lane change has been initiated. If it is determined that an inappropriate lane change is occurring, the Lane Change Assist 102 can, for example, also prevent a lane change to avoid dangerous driving behavior. The process is then completed in step S308.

[0054] Fig. Figure 4 shows the grid-like pattern of 400 discrete distance ranges. These distance ranges can be used, for example, in step S204 of the procedure, based on... Fig. The method described in section 2 allows the division of distances, which were detected in step S202, for example by the distance sensors of vehicles 106, into discrete distance ranges. The grid 400 can also be used in step S302 of the sub-method according to... Fig. The three measured distances are to be divided into discrete distance ranges. As a purely illustrative example, the average speed of the lane changes under consideration is 120 km / h, which corresponds to a recommended safety distance of 60 m. The average length of a vehicle (104) was assumed to be 4.36 m. The average width of a lane was assumed to be 3.5 m.

[0055] At the center of grid 400 is a vehicle 104 410 performing the lane change (hereinafter referred to as the ego vehicle). A longitudinal axis 402 of grid 400 has five distance ranges corresponding to the following longitudinal distances: more than 60 m in front of the ego vehicle (ahead), between 4 m and 60 m in front of the ego vehicle (directly in front), between 4 m in front and 4 m behind the ego vehicle (side-by-side), between 4 m and 60 m behind the ego vehicle (directly behind), and more than 60 m behind the ego vehicle (behind). A transverse axis 404 of grid 400 indicates the lane in which an object is located relative to the ego vehicle. The transverse axis has three distance ranges, which correspond to the following transverse distances: between 1.5 m and 4.5 m to the left of the ego vehicle (left), between 1.5 m to the left and 1.5 m to the right of the ego vehicle (same lane) and between 1.5 m and 4.5 m to the right of the ego vehicle 410 (right).

[0056] Fig. Figure 5 shows a schematic representation of five exemplary groups 500, 502, 504, 506, 508, which are the result of step S208 of the procedure according to Fig. The five groups 500, 502, 504, 506, and 508 are the exemplary result of the cluster analysis for a dataset of 104,000 lane changes. Each of these groups 500, 502, 504, 506, and 508 corresponds to a lane change scenario, which are described in more detail below.

[0057] The first group of 500 is designated as "free overtaking / free entry." The corresponding lane change scenario is characterized by a change to a lane to the left of the ego-vehicle 510. The lane changes of the first group of 500 involve overtaking vehicles that are traveling slower or at the same speed as the ego-vehicle 510 at the start of the lane change. During the lane change of the first group of 500, the ego-vehicle 510 accelerates in most cases. Occasionally, this results in exceeding the speed limit. The lane changes of the first group of 400 are performed with a small speed difference. The lane change scenario of the first group of 500 often begins in the rightmost lane.Such lane-changing scenarios frequently occur at points where the number of lanes increases or decreases, for example, at on- and off-ramps of highways and similar roads, and less often when exiting a construction zone. In many cases, a lane change of the first group 500 is initiated by activating the turn signal. A lane change of the first group 500 may be inappropriate if, for example, overtaking is prohibited, such as within a construction zone. This can be deduced, for instance, from the type of lane markings.

[0058] A second group 502 is designated as "Overtaking / Entering in Heavy Traffic." The corresponding lane-change scenario is characterized by heavy traffic, meaning that there are other vehicles in front of and behind the ego-vehicle 510, as well as to its left, that are traveling at or slower than the ego-vehicle 510 at the start of the lane change. During the lane change of the second group 502, the ego-vehicle 510 accelerates in most cases. Occasionally, this results in exceeding the speed limit. The speed during lane changes of the second group 502 is often slow. The lane-change scenario of the second group 502 often begins in the right-hand or a center lane. Just like the first group 500, lane-change scenarios of the second group 502 frequently occur at points where the number of lanes increases or decreases.A lane change in the second group 502 is also often initiated by activating the turn signal. A lane change in the second group 502 may be inappropriate if, for example, overtaking is prohibited or if traffic is too heavy, leaving the ego vehicle 510 with insufficient space to merge.

[0059] A third group, 504, is referred to as "free driving." The corresponding lane-change scenario is characterized by a rapid lane change in low traffic. Occasionally, lane changes in the third group, 504, involve exceeding the speed limit. The third group, 504, includes lane changes in which the ego-vehicle 510 accelerates, decelerates, or maintains its speed. The lane-change scenario of the third group, 504, begins equally in all lanes, but only rarely at on- or off-ramps of highways and similar roads. Lane changes in the third group, 504, are less frequently initiated by activating the turn signal. A lane change in the third group, 504, may be inappropriate if, for example, overtaking is prohibited.

[0060] A fourth group 506 is designated as "return due to the requirement to drive on the right." The corresponding lane change scenario is characterized by the ego-vehicle 510 merging into the right lane. The ego-vehicle 510 does not change its speed or slows down. In some cases, other vehicles are in front of the ego-vehicle 510 at the start of the lane change in the fourth group 506. Rarely, if ever, are other vehicles behind the ego-vehicle 510 at the start of the lane change in the fourth group 506. The lane change scenario of the fourth group 506 often begins in the left or a center lane, but only rarely at on- or off-ramps of highways and similar roads. Lane changes in the fourth group 506 are also frequently initiated by activating the turn signal.A lane change by the fourth group 506 may be inappropriate if the traffic volume is too high, so that the ego vehicle 510 has too little space when merging.

[0061] A fifth group, 508, is referred to as "making way." The corresponding lane-change scenario is characterized by the ego-vehicle 510 quickly moving into a right-hand lane to allow following vehicles to pass. The ego-vehicle 510 does not change its speed or slows down. The lane-change scenario of the fifth group, 508, often begins in the left or middle lane, but only rarely at on- or off-ramps of highways and similar roads. Lane changes of the fifth group, 508, are less frequently initiated by activating the turn signal. A lane change of the fifth group, 508, may be inappropriate if traffic is too heavy, leaving the ego-vehicle 510 with insufficient space to merge.

[0062] The five groups 500, 502, 504, 506, and 508 can be used not only to characterize the lane changes performed by vehicles 106 of fleet 108. All five groups 500, 502, 504, 506, and 508 are characterized by specific conditions at the beginning of the lane change. This classification is therefore also suitable for classifying the driving scenarios of vehicle 104, for example, to determine whether a lane change should be initiated and / or would be appropriate. This classification can be carried out, for example, by the classifier used in step S210 of the procedure according to Fig. 2 is being trained. Reference symbol list 100 System 102 Lane Change Assist 104, 106 Vehicle 108 vehicle fleet 110 Sensor 112 Vehicle system 114 Data transmission network 116 Sensor 118 Vehicle system 400 grid 402 Longitudinal axis 404 Transverse axis Group 500, 502, 504, 506, 508 510 Ego vehicle

Claims

[1] Method for operating a lane change assistant (102) of a vehicle (104) wherein a) Sensor data from sensors (110) of vehicles (106) of a vehicle fleet (108) are recorded, wherein at least part of the sensor data is recorded as a time series of data points; b) based on the sensor data, a large number of lane change data are generated, each of which is assigned to a lane change of one of the vehicles (106) of the vehicle fleet (108), wherein, from the part of the sensor data which has been recorded as a time series of data points, at least those data points which have been recorded at the beginning of the lane change and at the end of the lane change are used to generate the lane change data; c) Fleet data is generated based on lane change data; d) the fleet data are processed using cluster analysis to classify the lane changes into at least two groups (500, 502, 504, 506, 508) that differ in their lane change scenario; and e) is operated on the basis of cluster analysis of the lane change assistant (102). [2] Method according to claim 1, wherein at least one driving scenario of the vehicle (104) is classified on the basis of cluster analysis and the lane change assistant (102) is operated on the basis of the classification. [3] Method according to claim 2, wherein the classification determines whether the lane change assistant (102) is activated. [4] Method according to claim 2 or 3, wherein a decision is made on the basis of the classification as to whether a lane change should take place and the lane change assistant (102) is operated on the basis of this decision. [5] Method according to any one of claims 2 to 4, wherein, based on the classification, a decision is made as to whether a lane change is appropriate and the lane change assistant (102) is operated on the basis of this decision. [6] Method according to one of the preceding claims, wherein a machine learning method is trained on the basis of the result of the cluster analysis to generate a classifier which is trained to classify driving scenarios of the vehicle (104) on the basis of input data corresponding to sensor data of the vehicle (104) and / or quantities derived therefrom; and wherein the lane change assistant (102) is operated using the classifier. [7] Method according to claim 6, wherein the input data each comprise data from at least one of the following sensors (110) of the vehicle (104): a lateral acceleration sensor, a longitudinal acceleration sensor, a distance sensor, an external camera, an internal camera and a position sensor. [8] Method according to claim 6 or 7, wherein the input data each comprise the state of at least one of the following vehicle systems of the vehicle (104): a direction indicator and a driver assistance system, in particular the lane change assistant (102). [9] Method according to any one of claims 6 to 8, wherein the input data each comprise vehicle data corresponding to at least one vehicle information relating to the vehicle (104). [10] Method according to any one of claims 6 to 9, wherein the input data each comprise environmental data which corresponds to at least one environmental information relating to the environment of the vehicle (104). [11] Method according to any one of claims 6 to 10, wherein the input data each comprise driver data corresponding to at least one driver information relating to a driver of the vehicle (104). [12] Method according to any of the preceding claims, wherein the cluster analysis is performed as unsupervised learning. [13] Method according to any of the preceding claims, wherein the sensor data are continuously recorded and the lane change data and fleet data are regenerated and / or updated at regular intervals. [14] Method according to any of the preceding claims, wherein the sensor data and / or lane change data are anonymized before the fleet data are generated. [15] System (100) for operating a lane change assistant (102) of a vehicle (104), comprising a receiving unit (116) designed to receive sensor data from sensors (110) of vehicles (106) of a vehicle fleet (108); a processing unit (118) which is trained to generate a large number of lane change data based on the sensor data, each of which is assigned to a lane change of one of the vehicles (106) of the vehicle fleet (108), to generate fleet data based on the lane change data and to process the fleet data using cluster analysis in order to classify the lane changes into at least two groups (500, 502, 504, 506, 508) which differ with respect to their lane change scenario; and at least one lane change assistant (102) of a vehicle (104) which is trained to be operated on the basis of the result of the cluster analysis.

Images