Methods for predicting a dangerous driving situation

Abstract

Method for predicting a dangerous driving situation of a motor vehicle (2), wherein aspects relating to a driving situation of the motor vehicle (2) are recorded during a time interval, these aspects comprising operating parameters of the motor vehicle (2), environmental parameters of an environment in which the motor vehicle (2) is driving, and behavioral parameters relating to the activities of a driver (4) of the motor vehicle (2), wherein the behavioral parameters are used to determine whether the driver (4) is using a telephone and / or is distracted by using other devices for providing information and / or entertainment while driving, wherein the recorded aspects form a current pattern which is compared with known patterns during a time interval, and wherein at least one warning signal is provided to the driver (4) if the current pattern corresponds to a known pattern.which indicates a dangerous driving situation with a defined probability, wherein at least one known pattern indicating a dangerous driving situation has been recorded in the past during a time interval during an actual journey of at least one other motor vehicle, wherein this time interval ended with a dangerous driving situation.

Description

The invention relates to a method for predicting a dangerous driving situation of a motor vehicle and to a system for predicting a dangerous driving situation of a motor vehicle. Typically, a sensor-based system, such as the emergency braking system, is used to prevent a motor vehicle accident. This system, installed in the vehicle, detects an approaching object and uses the vehicle's speed to determine whether a collision is imminent. If the risk of a collision is deemed high, the vehicle initiates emergency braking. Another system provides a break recommendation. If increasing driver fatigue is detected, the system advises the driver to take a break. German patent application DE 10 2009 011 259 A1 describes a vehicle-to-vehicle position detection system that uses wireless information transmission techniques. One embodiment of the system comprises a detection and distance measurement system mounted on a host vehicle, the detection and distance measurement system being configured to detect a neighboring vehicle in the vicinity of the host vehicle. Upon detection of the neighboring vehicle, the detection and distance measurement system generates neighboring vehicle data indicating the position of the neighboring vehicle relative to the host vehicle. The position detection system also includes a traffic modeling device configured to process the neighboring vehicle data and, in response, generate a virtual traffic model for the host vehicle.The positioning system also employs a wireless transmitter that wirelessly transmits host vehicle model data, which in turn transmits the virtual traffic model. Compatible vehicles in close proximity to the host vehicle can receive and process this data to generate their own virtual traffic models. Documents US 2007 / 0043502A1 and US 2012 / 0306663A1 each disclose a method in which motor vehicles communicate with each other to predict an impending collision. A method for controlling safety devices in a motor vehicle during a driving situation is described in German patent application DE 10 2007 039 038 A1. The aim is to prevent a collision between the motor vehicle and an object in the surrounding environment. Furthermore, motion data of at least one object in the surrounding environment is determined. From this motion data, a situational hypothesis regarding the at least one object in the surrounding environment is derived. In addition, an assessment of potential hazards in the current driving situation is carried out, and the driver is alerted to any such hazard. A method for traffic-guided control and / or support of motor vehicles is known from German patent application DE 199 38 691 A1. In this method, the area surrounding the motor vehicle is monitored by distance sensors. Based on this data, a traffic situation is assessed. If there is an imminent danger to the motor vehicle, a driver assistance system activates the accelerator or brake pedal. A method for the prognostic evaluation of at least one predictive safety system of a motor vehicle, taking into account statistical real accident data describing several accidents and stored in a database, is known from publication DE 10 2008 027 509 A1. Against this background, a method and a system with the features of the independent patent claims are presented. Embodiments of the method and the system are described in the dependent patent claims. The method according to the invention is designed to predict a dangerous driving situation of a motor vehicle. During the execution of the method, aspects relating to and / or describing a driving situation of the motor vehicle are recorded over a time interval. These aspects include operating parameters of the motor vehicle, environmental parameters of the environment in which the motor vehicle is driving, and behavioral parameters relating to and / or describing the actions of the driver of the motor vehicle. Furthermore, the recorded aspects are designed to form a current pattern, which is compared with known patterns at regular intervals. The driver is provided with at least one warning signal if the current pattern corresponds to a known pattern that indicates the dangerous driving situation with a defined probability. The patterns are typically recognized automatically during a learning or training phase of the system by at least one algorithm, such as a machine learning algorithm for computer-aided machine learning. Depending on the algorithm used, the patterns may or may not be human-readable. An artificial neural network used in this context, for example, can be described as a black box, since the patterns underlying the decision are not apparent. In practice, different algorithms are tested during the training phase, which allows the patterns formed from the various aspects to be classified. The detected aspects are compared with each other and / or with each other at intervals, whereby at least one warning signal is provided to the driver if the detected aspects form a known pattern that usually indicates the dangerous driving situation with a defined probability. Typically, a pattern identified from the recorded aspects is compared with known patterns that lead to a dangerous driving situation. If the probability that the currently detected pattern leads to a dangerous driving situation is at least as high as the definable or to-be-defined value or target value for a known pattern, at least one warning signal is triggered and thus made available. The aforementioned probabilities are typically determined using a machine learning algorithm. A variety of different methods are conceivable. The most suitable method is determined through testing during the learning phase of at least one algorithm. The probabilities are automatically calculated based on comparable historical situations. An evaluation of data relating to such historical situations is performed by at least one algorithm. By providing at least one visual and / or acoustic warning signal, it is possible to avoid dangerous situations, e.g., an accident. Patterns indicating a dangerous driving situation are specified in the procedure from a database. At least one such known pattern, and generally every known pattern indicating a dangerous driving situation, must have been recorded and / or captured in the past during a time interval during an actual journey by at least one other motor vehicle, with this time interval ending with a dangerous driving situation. Such a known pattern from the past must have been recorded by an electronic and / or computer-based tachograph or data recording device of the at least one other motor vehicle involved in the dangerous driving situation.This data recording device is typically connected to all control units and sensors within the vehicle via a network and is centrally located for data exchange between these units and sensors. All data used and received by the control units and sensors is captured and / or read by the data recording device. This device is, for example, configured as a gateway and linked as a node in the network to all control units and / or sensors. The "at least one aspect" typically describes whether a specific event can take place and / or occur during the time interval, for example, whether a device in the vehicle is switched on and / or used by the driver. Thus, it is necessary to determine, among other things, whether the driver made a phone call during the time interval. In this embodiment, at least one aspect describes a parameter, whereby actual values ​​of parameters are recorded as aspects during the time interval. These parameters include operating parameters of the vehicle, environmental parameters of the environment in which the vehicle is driving, and behavioral parameters describing the activities of the vehicle's driver. Each actual value recorded during the time interval is compared with at least one designated target value for the parameter. At least one warning signal is provided to the driver if a combination of actual values ​​of several parameters deviates from a designated target value by at least one tolerance value. The current pattern is typically described qualitatively and / or quantitatively based on its current aspects. Each aspect is also described qualitatively and / or quantitatively based on a parameter. For example, an aspect may be quantitatively derived from the current actual value of a parameter. If the only consideration is whether the actual value of the parameter is greater or less than the target value, at least a qualitative description is required. This description serves to determine whether a device is switched on or not. The time interval is usually only a fraction of a second, at most a few seconds, and is always shorter than the time interval, which can last several dozen seconds, e.g. about 1 minute. Taking these aspects into account, a driving situation can be evaluated qualitatively and / or quantitatively. The system according to the invention is designed to predict a dangerous driving situation of a motor vehicle, wherein the system comprises at least one sensor, at least one control device, and at least one warning system. The at least one sensor is designed to detect, during a time interval, aspects relating to and / or describing a driving situation of the motor vehicle, wherein these aspects include operating parameters of the motor vehicle, environmental parameters of an environment in which the motor vehicle is driving, and behavioral parameters relating to and / or describing the activities of a driver of the motor vehicle. The system is designed so that the recorded aspects form a current pattern. The at least one control device is configured to compare the current pattern with known patterns at regular intervals, and the at least one warning system is configured to provide the driver with at least one warning signal if the current pattern corresponds to a known pattern that indicates a dangerous driving situation with a definable probability. The at least one control device is designed to compare the recorded aspects with each other and / or with each other during a time interval. The at least one warning system is additionally designed to provide the driver with at least one warning signal if the recorded aspects form a pattern that indicates a dangerous driving situation with a defined probability. The system comprises a network that connects at least one sensor and at least one control device and is located in the motor vehicle. Furthermore, the system has at least one database in which known patterns, each indicating a dangerous driving situation, are stored and / or will be stored. The at least one database is stored in a stationary, central data processing system comprising a computing unit. Alternatively or additionally, the at least one database is stored in a computing unit within the vehicle and integrated into the network. Information relating to known patterns shall be transferred from the database in the stationary, central data processing system to the database in the control device of the motor vehicle, whereby the database in the control device of the motor vehicle shall be supplemented and / or updated with regard to new information relating to the known patterns. In a further embodiment of the system, at least one aspect describes a parameter. The at least one sensor is configured to capture actual values ​​of parameters as aspects during the time interval. These parameters include operating parameters of the vehicle, environmental parameters of the environment in which the vehicle is operating, and behavioral parameters describing the actions of the vehicle's driver. The at least one control device is configured to compare each captured actual value during the time interval with at least one designated target value for the parameter. Furthermore, the at least one warning system is configured to provide the driver with at least one warning signal if a combination of actual values ​​of several parameters deviates from a designated target value by at least one tolerance value. The method and system can predict and avoid a potentially dangerous driving situation, which can usually end in an accident, by analyzing a pattern of signals that occur during a journey of the motor vehicle in the network of interconnected control devices of the motor vehicle (e.g., designed as a bus system) and describe an operating state of the motor vehicle, by issuing at least one warning signal. The procedure is typically implemented in two phases. For the first phase, a learning phase, an electronic tachograph configured as a data recording device or logger is provided in at least one other vehicle. Such a data recording device can be implemented as a function and / or module of the at least one, usually already existing, control unit and / or installed in the at least one other vehicle. This data recording device is configured to continuously store or persist all information from signals and / or data from the entire bus system, encompassing all control units, in a non-volatile storage medium as patterned aspects, using data flow control or a so-called sliding window approach, for a time interval of, for example, 60 seconds before each current moment.In the event of an accident, the data collected during the preceding time interval, containing the relevant information, is anonymized and transmitted to a central service provider and / or vehicle manufacturer, such as Audi, where it is compiled in a continuously updated database or data repository. To obtain and / or analyze reference situations in which no accidents occurred, the service provider and / or vehicle manufacturer also collects test data. After a project period, the database or data basis comprises a very large number of information packages, which cover the aspects and are based on data and / or signal patterns from a large number of centrally recorded time intervals, e.g., each 60 seconds long, from a large number of motor vehicles, whereby each information package is marked or labeled with a label "accident" if it was recorded in a time interval before an accident involving a motor vehicle, or with the label "no accident" if it was determined during an accident-free journey. Data derived from such information packets undergoes preprocessing to identify relevant patterns of aspects, such as actual values ​​of vehicle operating parameters. This allows for the differentiation between patterns that lead to or result from an accident and reduces data complexity. Based on this preprocessed data, at least one algorithm—a classification algorithm and / or regression algorithm—is trained. This algorithm can distinguish between an accident and the absence of an accident based on a given pattern of aspects. Various algorithms are used for this purpose, either performing a binary classification or outputting a probability for a specific group. As part of the process, aspects are automatically determined during the learning phase or learning process, for example, by selecting at least one algorithm, distinguishing between aspects that indicate accidents and those that do not. The selection of relevant aspects, usually parameters, from the total number of aspects is part of the automatic preprocessing and is commonly referred to as "feature selection" or "dimensionality reduction." This makes it possible to use at least one algorithm designed as a classification algorithm and / or as a regression algorithm. In this context, a classification algorithm is referred to as a "random forest," which uses uncorrelated decision trees that grow as the algorithm is executed. These decision trees can include hierarchically structured aspects, and a decision is made as to whether a particular combination of aspects within such a decision tree has a probability of a dangerous driving situation or not. A "classification tree" or classification tree method is at least one additional classification algorithm. Here, the intended aspects are used as classifications that are combined with each other, with certain combinations indicating a dangerous driving situation and other combinations indicating otherwise. A "Support Vector Machine" is an algorithm, also known as a support vector machine, in which aspects are categorized into classes. Collections and / or combinations of aspects within a given class form patterns that can be used to determine whether or not they indicate a dangerous driving situation. A Support Vector Machine can be used as a classification algorithm and / or as a regression algorithm. An alternative or supplementary classification algorithm is the so-called nearest neighbor (K-Nearest Neighbor) classification. This also combines aspects into patterns, taking into account how certain aspects, as objects, relate to other aspects and are adjacent to each other. In another classification algorithm, dynamic models are formed from aspects, taking into account a temporal sequence of the aspects in order to form and / or recognize patterns encompassing these aspects. In an algorithm designed as a regression algorithm, a neural network is used, forming patterns from aspects. Deep learning, as a usable algorithm, checks combinations of aspects to see if they form patterns. In this case, too, patterns of aspects are sought, and a decision is made as to whether these could cause a dangerous driving situation or not. Furthermore, it is possible to use a generalized linear regression as the regression algorithm. Here, aspects can be used as variables and examined to determine the extent to which they form systematic patterns, allowing a statement to be made as to whether patterns involving combinations of aspects lead to dangerous driving situations or not. The network data is monitored by the data recording device (gateway). This device contains virtually all data signals exchanged between control units and / or sensors during vehicle operation. A feature selection algorithm then uses this data to identify the information that best differentiates between the two scenarios: "accident" and "no accident." After successful analysis and / or learning of the patterns from aspects, the regression and / or classification algorithm based thereon is integrated into at least one control device or vehicle computer. This device uses information from a past time interval, e.g., during the last 60 seconds, which is currently generated in the network or bus system of the motor vehicle at a predetermined time interval, e.g., every 100 ms, to control and / or monitor at least one control device in order to analyze and classify this current, driving-related information based on the patterns from aspects.If, as a rule, a pattern of aspects is recognized by comparing the current pattern with known patterns, and this pattern is usually recognized again, and this pattern has led to an accident with a high probability in a previously learned case, the driver of the motor vehicle is warned visually and / or audibly by at least one warning signal. In its implementation, the procedure allows for the acquisition of operating parameters of various devices and / or equipment, as well as environmental conditions, via sensors during a journey. Based on this data, events and thus aspects can be classified. The pattern that characterizes the current driving situation of the vehicle can be derived from the type and / or frequency of such aspects. An example driving situation can be characterized by the following aspects and / or events: A driver of a motor vehicle is approaching a tight bend (aspect 2) with the windshield defroster activated at an outside temperature of -5°C (aspect 1) and is simultaneously talking on the phone with his wife (aspect 3). He is also currently adjusting the seat heating (aspect 4) and the high beam assist (aspect 5). Furthermore, an oncoming vehicle is detected (aspect 6). Additionally, the lane departure warning system has intervened twice in the last 60 seconds (aspect 7). Based on known patterns for similar driving situations in the past and due to a cluster of aspects indicating a dangerous driving situation, the system has determined an accident probability of 98% and therefore issues a warning signal to the driver. The presented system and procedure take into account at least all aspects of the vehicle, the driver, and the surroundings / environment that are defined as parameters in order to detect an impending accident at an early stage. Thus, the interplay of the vehicle and driver factors, which typically interact with each other, is considered together. This allows for the early detection of a critical driving situation, whereas an emergency braking system only detects such a situation once the vehicle is already in a dangerous situation. Alternatively or additionally, at least one intervention algorithm must be considered within the procedure. This algorithm is automatically defined, taking different driving situations into account. This allows accidents to be detected early enough to anticipate dangerous driving situations. As part of the process, signals containing information about the aspects are examined in the vehicle's bus system for the presence of hidden patterns of aspects that differ from each other in every driving situation. Using at least one self-learning algorithm, the probability of an accident occurring in a future time interval, i.e., within the next x seconds, is classified based on these patterns of aspects. The procedure and the system allow for the timely detection of dangerous driving situations that could result in an accident. The at least one warning signal for the driver helps to prevent the potential accident before it occurs. Further advantages and embodiments of the invention will become apparent from the description and the accompanying drawing. It is understood that the features mentioned above and those to be explained below can be used not only in the combinations specified, but also in other combinations or on their own, without leaving the scope of the present invention. The invention is schematically illustrated with reference to embodiments in the drawing and is described schematically and in detail with reference to the drawing. Fig. 1 shows a schematic representation of details of an embodiment of the system according to the invention. Fig. 1 shows a schematic representation of a motor vehicle 2, driven by a driver 4. This motor vehicle 2 comprises a control unit 6, which in turn includes a processing unit 7 and a memory 8 on which a vehicle-internal database is stored. The motor vehicle 2 also includes several sensors 10, 12, 14. In this case, at least one first sensor 10 is designed to detect actual values ​​of operating parameters of the motor vehicle 2 as aspects within the framework of an embodiment of the method according to the invention. At least one second sensor 12 is arranged on an outer wall of the motor vehicle 2 and, in the embodiment of the method, is configured to detect actual values ​​of environmental parameters of the environment in which the motor vehicle 2 is currently located. Such environmental parameters are typically meteorological parameters and include information about the current weather conditions. These environmental parameters, and thus meteorological parameters, are temperature, air pressure, and humidity. At least one meteorological parameter also indicates whether precipitation is currently occurring. Furthermore, the at least one second sensor 12 can also detect the brightness or darkness of the environment through which the motor vehicle 2 is traveling. The at least one third sensor 14 is designed to detect aspects of the driver's behavior and / or activities within the framework of the embodiment of the method. For example, this at least one third sensor 14 can detect whether the driver 4 is using a telephone while driving and / or might be distracted by using other devices, such as those used for providing information and / or entertainment. Furthermore, the at least one third sensor 14 can detect the driver 4's current biological and / or medical condition. Thus, it is possible, for example, to determine whether the driver 4 is possibly fatigued. These sensors 10, 12, 14 and the at least one control unit 6 are connected to each other via communication lines in a network 16 of the motor vehicle 2, which is also referred to as a bus system, whereby it is provided that actual values ​​of parameters and thus of aspects that are recorded by the sensors 10, 12, 14 are transmitted to the control unit 6 and evaluated there. Furthermore, Fig. 1 shows a central, stationary data processing system 18 on which a stationary, central database is stored. Fig. 1 also shows a communication antenna 20 assigned to the motor vehicle 2, via which the at least one control device 6 can communicate with the data processing system 18 and exchange data. The aforementioned components of the motor vehicle 2, which simultaneously form the network 16 of the motor vehicle 2, as well as the central stationary data processing system 18, are designed as components of the embodiment of a system 22 according to the invention. Furthermore, this system 22 comprises at least one warning system 24 arranged in the motor vehicle 2. In the embodiment of the method, it is provided that during the journey of the motor vehicle 2, actual values ​​for the parameters and thus for the aspects are continuously recorded via the aforementioned sensors 10, 12, 14 and transmitted to the computing unit 7. The system is designed to derive a current pattern from continuously recorded aspects during a given time interval and compare it with previously known patterns for those aspects. These previously known patterns for aspects stored in the vehicle's internal database indicate a dangerous driving situation. As soon as a current pattern from currently recorded driving-related aspects during a given time interval indicates a previously known pattern, the driver is visually and / or audibly alerted to the impending dangerous driving situation by at least one signaling device 24 of the system 22. In the described embodiment of the method, it is further provided that such known patterns for aspects indicating a dangerous driving situation are also stored in the central stationary database of the central stationary data processing system 18. These known patterns are provided to the control unit 6 via communication based on electromagnetic waves, e.g., using the internet, via the communication antenna 20. Furthermore, the currently determined known patterns for aspects from the central stationary database update and thus continuously renew previously known patterns for aspects in the database of the control device 6 and therefore of the motor vehicle 2. Furthermore, the embodiment of the method provides that patterns relating to aspects stored in the central stationary database of the central stationary data processing system 18 are previously provided to the central database by other motor vehicles. In this regard, it is also provided that each such other motor vehicle, as a participant in a motor vehicle fleet, also has sensors like the motor vehicle 2 explicitly depicted in Fig. 1, with which aspects and thus parameters of the respective other motor vehicle are also recorded during the journey and stored for a period of time. Should it transpire that a motor vehicle of this motor vehicle fleet enters a dangerous driving situation, orIf a dangerous driving situation has occurred, the pattern of those aspects that were recorded during the time interval before the dangerous driving situation is automatically transmitted to the data processing system 18, evaluated by the data processing system 18 and the existing central, stationary database is updated on the basis of such a pattern.

Claims

Method for predicting a dangerous driving situation of a motor vehicle (2), wherein aspects relating to a driving situation of the motor vehicle (2) are recorded during a time interval, these aspects comprising operating parameters of the motor vehicle (2), environmental parameters of an environment in which the motor vehicle (2) is driving, and behavioral parameters relating to the activities of a driver (4) of the motor vehicle (2), wherein the behavioral parameters are used to determine whether the driver (4) is using a telephone and / or is distracted by using other devices for providing information and / or entertainment while driving, wherein the recorded aspects form a current pattern which is compared with known patterns during a time interval, and wherein at least one warning signal is provided to the driver (4) if the current pattern corresponds to a known pattern.which indicates a dangerous driving situation with a defined probability, wherein at least one known pattern indicating a dangerous driving situation has been recorded in the past during a time interval during an actual journey of at least one other motor vehicle, wherein this time interval ended with a dangerous driving situation. The method according to claim 1, wherein known patterns, each indicating a dangerous driving situation, are specified from at least one database. Method according to claim 1 or 2, wherein at least one aspect describes whether a certain event takes place during the time interval. A method according to one of the preceding claims, wherein at least one aspect describes a parameter, wherein actual values ​​of parameters are recorded as aspects during the time interval, wherein these parameters include operating parameters of the motor vehicle (2), environmental parameters of an environment in which the motor vehicle (2) is driving, and behavioral parameters describing the activities of the driver (4) of the motor vehicle (2), wherein each recorded actual value is compared during the time interval with at least one target value provided for the parameter, wherein at least one warning signal is provided for the driver (4) if a combination of actual values ​​of several parameters deviates from a target value provided for this purpose by at least one tolerance value. Method according to one of the preceding claims, wherein the time step is shorter than the time interval. System for predicting a dangerous driving situation of a motor vehicle (2), wherein the system (22) comprises at least one sensor (10, 12, 14), at least one control device (6) and at least one warning device (24), wherein the at least one sensor (10, 12, 14) is configured to detect aspects relating to a driving situation of the motor vehicle (2) during a time interval, wherein these aspects include operating parameters of the motor vehicle (2), environmental parameters of an environment in which the motor vehicle (2) is driving, and behavioral parameters relating to the activities of a driver (4) of the motor vehicle (2), wherein the behavioral parameters are to demonstrate whether the driver (4) is using a telephone and / or is distracted by the use of other devices for providing information and / or entertainment while driving, wherein the detected aspects form a current pattern, and wherein the at least one control device (6) is configured toto compare the current pattern with known patterns during a time interval, wherein the at least one warning device (24) is configured to provide at least one warning signal to the driver (4) when the current pattern corresponds to a known pattern that indicates a dangerous driving situation with a defined probability, wherein at least one known pattern indicating a dangerous driving situation has been detected in the past during a time interval during an actual journey of at least one other motor vehicle, wherein this time interval ended with a dangerous driving situation. System according to claim 6, comprising a network (16) that connects the at least one sensor (10, 12, 14) and the at least one control device (6). System according to claim 6 or 7, comprising at least one database in which known patterns, each indicating a dangerous driving situation, are stored. System according to claim 8, wherein the at least one database is stored in a stationary, central data processing system (18). System according to claim 8 or 9, wherein the at least one database is stored in the control device (6) of the motor vehicle (2). System according to claims 9 and 10, wherein information relating to patterns is to be transmitted from the database in the stationary, central data processing system (18) to the database in the control device (6) of the motor vehicle (2), wherein the database in the control device (6) of the motor vehicle (2) is to be supplemented with regard to the information relating to the known patterns. System according to one of claims 6 to 11, wherein at least one aspect describes a parameter, the at least one sensor (10, 12, 14) being configured to detect actual values ​​of parameters as aspects during the time interval, wherein these parameters include operating parameters of the motor vehicle (2), environmental parameters of an environment in which the motor vehicle (2) is driving, and behavioral parameters describing the activities of the driver (4) of the motor vehicle (2), wherein the at least one control device (6) being configured to compare each detected actual value during the time interval with at least one target value provided for the parameter, wherein the at least one warning system (24) being configured to provide the driver (4) with at least one warning signal when a combination of actual values ​​of several parameters deviates from a target value provided for this purpose by at least one tolerance value.

Images