EP4246487A1 - Methods and systems for early detection and assessment of structural hazardous locations in road traffic - Google Patents

Abstract

The invention relates to a system or method for the early detection of structural danger spots in road traffic using a digital traffic route network image or map. For this purpose, accident data (1), user input data (3) and sensor data (2), according to their georeferencing, are assigned to segments of the map (500) using a computer system (5). According to one aspect, the computer system (5) carries out an evaluation, comprising at least a frequency determination and/or a data comparison. For the purpose of early detection, a georeferenced segment to which no accident data (1) is assigned is identified as a potential danger point if sensor data (2) or user input data (3) have been assigned with a predetermined frequency and/or if a data comparison, in particular a comparison, has been carried out with an AI-supported model, user input data (3) and/or sensor data (2) shows that there is a predefined degree of feature agreement or feature correlation with critical patterns. This enables early detection of danger spots. According to a further aspect, it is proposed to determine a danger score for identified danger spots.

Description

TECHNICAL FIELD

The invention generally relates to methods and systems for the early detection and evaluation of structural danger points in road traffic. The invention relates in particular to a computer-aided method for the early detection of structural danger spots in road traffic using a digital traffic route network map using a computer system. The invention further relates to a computer-aided method for determining a danger score of a georeferenced, structural danger point in a digital traffic route network image, as well as a computer system for implementing the aforementioned methods and / or for using or making usable the map-like digital traffic route network image resulting from one of these methods , which technically represents a danger map. Furthermore, new uses of the computer system or the hazard map data are suggested. The invention also relates, among other things, to a method or a system for communication with autonomous or semi-autonomous driving vehicles in order to influence the autonomous or semi-autonomous driving behavior depending on danger spots.

In this area, a distinction is made between permanent or structural hazards and only situational, temporary hazards. Permanent danger spots are always present. Temporary danger spots, on the other hand, are limited in time (such as at construction sites) or they only occur in connection with certain general conditions, such as adverse weather conditions (be it in the summer due to the glare of the sun, in the rain due to aquaplaning or in the winter due to snow and ice). . There are established solutions for reporting temporary danger spots, such as RDS/TMC according to ISO 14819. Early detection is, of course, hardly possible here.

The present invention relates at least predominantly or exclusively to permanent or structural danger points in the road network. In particular, the term danger zone is not to be understood as limited to danger to a specific type of road user, i.e. it includes danger to different types of road user. A danger point can also be understood as a risk point.

BACKGROUND OF THE INVENTION

Since the number of fatalities in traffic accidents has been stagnating for a long time and no further decline in fatal accidents is expected and at the same time the number of cycling accidents, especially involving e-bikes, is increasing sharply, research is being carried out intensively into how traffic route networks can be made safer. For some time now, computer-aided methods for identifying danger spots on these traffic route networks have been increasingly being tested and used to tentatively predict accident hazard spots for motor vehicles, typically based on statistical methods or probability calculations, and thus to defuse the corresponding point on the traffic route network before an accident occurs.

STATE OF THE ART

Known methods are based, for example, on collecting and mapping known accident data from accident databases of, for example, police authorities, and on determining where a danger zone exists based on the frequency of accidents. With this approach, dangerous areas cannot be identified before accidents, some of which are serious, have already occurred. In addition, minor cases or minor accidents without a police report are typically not included in the accident data recorded by the authorities.

There is also the approach of reading sensor data or so-called pulse data through the vehicle electronics and/or additional instruments and using them to detect critical driving maneuvers. Such impulse data is now used by some insurance companies and/or vehicle fleet operators to check the quality of the driver's driving and thus be able to adjust any premiums. Impulse data can offer an opportunity to identify danger spots through critical driving maneuvers before accidents even occur. Dangerous areas identified in this way can then be reported to the road traffic authorities, for example, so that they can initiate the defusing of the dangerous areas.

However, practical application has shown that pulse data as the sole criterion is often not sufficient to accurately identify or predict danger spots. This is proven, among other things, by the results of a study from 2019, cf. B. Ryder, A. Dahlinger et al.; Spatial prediction of traffic accidents with critical driving events - Insights from a nationwide field study; Transportation Research Part A: Policy and Practice; Vol. 124, 2019 (pp. 611-626), ISSN 0965-8564 . This publication shows that there is a causal connection between impulse data and danger zones, but also makes it clear that impulse data alone is not sufficient to make reliable or accurate statements about danger zones.

Not all pulse data can be traced back to danger spots. Critical driving maneuvers that are included in the impulse data can be caused by various other causes, e.g. by inattentive drivers, so that there is no connection to the design of a traffic route. Furthermore, in practice it is difficult to detect accumulations of impulse data at certain points in the traffic network, especially if the accumulations are due to infrastructure but without a significant danger point, e.g. braking in front of traffic lights. Even road users with risky and/or inexperienced driving behavior can trigger false information about danger spots when using approaches that rely on sensor data or impulse data, since there does not have to be a direct connection between driving behavior and an actual danger spot.

A method for identifying potential danger spots in road traffic using a central computer unit using a fleet of networked motor vehicles is described in the DE 10 2020 108 531 A1 (Daimler AG ). Here, an event that indicates a potential danger point is recorded via pulse data and transmitted with its geoposition to the central processing unit. However, the aforementioned difficulties can only be partially overcome and require increased computing effort in the vehicles or in the central computer unit. In addition, this solution only applies to motor vehicles.

Furthermore, a procedure for warning of dangers in road traffic and a driving assistance system for warning in the DE 10 2019 203 405 A1 (Audi AG ). It is also mentioned here that data from various data sources relating to potential danger points should be used for the complex assessment of a potential danger point. Statistical evaluation is mentioned, but how the complex evaluation should be carried out is not described in more detail. In this respect, this still exists Need for a solution for the detection, especially early detection, of danger spots. The system off DE 10 2019 203 405 A1 focuses on dangers that exist for the driver of the motor vehicle and could only trigger a warning for him, and therefore does not allow a holistic view or assessment from the perspective of other road users.

A proposal for the complex assessment of danger zones is described on the project homepage of the International Road Assessment Program (iRAP), in particular the publication "iRAP Methodology Fact Sheet #6" available there. In the iRAP project, the assessment of a danger spot is calculated for each 100 m long road segment and for each type of road user.

The factors that are included in the calculation according to iRAP are road risk factors, which indicate how likely an accident is, road risk factors, which indicate how serious an accident would be, and speed factors, which indicate how the probability of an accident depends on the speed driven , external factors that indicate how likely it is to be involved in an accident involving third parties and a factor that takes into account the probability that a vehicle, for example from oncoming traffic, will be steered onto your own lane. These factors are then determined for each accident type and the result of the individual accident types are summed up per road segment and road user type, so that a rating is calculated for all accident types for a segment and a road user type. This approach, which is based on probability calculations, has very limited use at best for the early detection of danger spots. This approach also only aims at the automotive perspective.

The WO 2017 146 790 A1 (Allstate Insurance Co. ) in turn describes a system for generating a risk road map, which makes it possible to determine a risk for planned routes.

Previous approaches do not allow for sufficiently accurate early detection of danger spots, which results in, among other things, an unacceptable number of false reports of danger spots, which is prohibitive for practical usability.

Furthermore, in most cases only the dangers for motorized road users and/or danger spots are recorded from the perspective of these road users, so that danger spots for other road users or roads that are primarily for Different traffic routes, such as cycle paths or pedestrian paths, are not recorded for motor vehicles. Typically, weaker, non-motorized road users also lack a view of potential conflicts.

Another problem is that the traffic route network is only processed incompletely and usually only roughly, broken down into route sections, so that many actual danger spots remain undetected.

Previous approaches also often have the problem that little or no information is available about the type of danger point and only shows where a danger point is located.

TECHNICAL TASK

An object of the present invention is therefore to at least partially overcome known disadvantages or shortcomings from the prior art and, for this purpose, in particular to propose improved computer-aided methods and systems for early detection on the one hand and for accurate assessment of danger spots on the other hand. This object is achieved on the one hand by a method according to claim 1 or, independently thereof, on the other hand by a method according to claim 2.

Furthermore, a computer system or a digital hazard map for a computer system according to claim 14 is proposed.

Furthermore, for example, according to claim 15, a method for generating hazard-related warning messages to road users using a digital hazard map is proposed.

Further uses of the digital danger maps or of computer systems that have such a danger map are proposed in claim 16, in particular in connection with autonomous or semi-autonomous driving vehicles, for influencing autonomous or semi-autonomous driving behavior using the digital danger maps.

In particular, the procedures and systems should be able to identify structural hazards at an early stage, even before they become apparent through official accident statistics. Furthermore, an object of the present invention can be seen in proposing solutions that allow the transport network to be as optimal as possible largely or comprehensively, ideally covering the entire transport network.

Another independent task is to evaluate the traffic situation or the traffic route network as holistically as possible from different perspectives with regard to danger spots, i.e. to evaluate it from the different perspectives of different or all types of road users. In contrast to the prior art, which often carries out an evaluation from the perspective of only one type of participant, e.g. for motor vehicles, the invention should also make it possible to identify danger spots where different types of participants, especially without a vehicle involved, meet each other, e.g. where there is often a cyclist and pedestrians collide, or even identify risk areas where often only one type of road user has an accident (e.g. due to frequent wet leaves in one place or the like).

The procedures or systems should be usable for assessing dangers from the perspective and for the benefit of as many types of road users as possible, including weaker non-motorized groups such as cyclists or pedestrians.

GENERAL DESCRIPTION OF THE INVENTION

According to a first independent aspect, a computer-aided method for the early detection of structural danger spots in road traffic or a system set up for this purpose is proposed.

Here, a digital traffic route network map is processed using a computer system, for which, for example, map data from the OpenStreetMap (OSM) project or other comparable map data can be used.

The system or method initially includes providing such a digital map-like image of at least one traffic route network, in particular with infrastructure data and possibly traffic route metadata, whereby the map-like image can be or is loaded into a computer memory of the computer system.

The computer system first divides the map-like image into a large number of georeferenced segments of traffic routes, whereby the segments can preferably have a variable length for the purpose of data economy and/or more precise localization of danger spots.

According to a core idea, several different data sources, in particular three different data sources are used, the data of which are appropriately blended by the process or system. Particular preference is given to using digitized georeferenced accident data, georeferenced user input data and georeferenced sensor data that relate to movement characteristics of road users, in particular acceleration values.

1. a) zumindest initiales Bereitstellen von digitalisierten georeferenzierten Unfalldaten, umfassend statistische Unfalldaten, insbesondere in Form zumindest einer Datenbank, welche durch das Rechnersystem zugreifbar bzw. verarbeitbar sind;
2. b) Datenerheben von georeferenzierten Nutzereingabedaten bezogen auf potentielle Gefahrenstellen, wobei die Datenerhebung direkt oder durch das Rechnersystem, oder aber indirekt durch weitere Systeme bzw. Server erfolgen kann, welche die Daten dem Rechnersystem zur Verfügung stellen; und
3. c) Datenerfassen, insbesondere durch das Rechnersystem, von georeferenzierten Sensordaten bezogen auf Bewegungsmerkmale von Verkehrsteilnehmern, insbesondere basierend auf Beschleunigungsmesswerten, welche bewegungsbezogen von Sensoren generierbar sind bzw. generiert werden, die von Verkehrsteilnehmern oder Fahrzeugen mitgeführt werden, insbesondere von fahrzeuginternen oder geräteinternen Sensoren, z.B. in an sich bekannten Geräten wie Smartphones, Navigationssystemen, TelematikAusrüstung, loT-Geräten, usw.;

For this purpose, the invention provides in particular the following steps:

1. a) at least initially providing digitized georeferenced accident data, comprising statistical accident data, in particular in the form of at least one database, which can be accessed or processed by the computer system;
2. b) data collection of geo-referenced user input data related to potential danger spots, whereby the data collection can take place directly or through the computer system, or indirectly through other systems or servers that make the data available to the computer system; and
3. c) Data acquisition, in particular by the computer system, of geo-referenced sensor data related to movement characteristics of road users, in particular based on acceleration measurements which can be generated or are generated in relation to movement by sensors that are carried by road users or vehicles, in particular by vehicle-internal or device-internal sensors, e.g in well-known devices such as smartphones, navigation systems, telematics equipment, LoT devices, etc.;

The temporal frequency or frequency of data acquisition from steps a) to c) is typically different. The georeferenced data from the different data sources a), b) and c) can therefore be obtained in the form of separate data streams, at different times and, if necessary, at different time intervals. In particular, no real-time data collection is required, although possible, especially for sensor data c). If necessary, all data can be standardized by pre-processing the raw data. In addition to georeferencing, e.g. GPS coordinates, all or some data can have further information, such as a time binding or a timestamp.

Statistical accident data for a) can come from different sources, e.g. from police records, from insurance company records and/or from Emergency rooms/hospitals.

User input data for b) can first be collected by a server, e.g. a web server, and fed to the computer system through updates. An ongoing survey is not necessary; short update intervals are desirable. The data collection can be initiated by a user himself and/or by the computer system, for example by querying. User input data can in particular be collected cumulatively or incrementally. User input data can also be fed directly to the computer system.

The acquisition of sensor data for c) (also called pulse data) is carried out in particular, but not necessarily, with short update intervals, but not in real time, in order to enable the most up-to-date early detection and mapping of danger spots without significant communication effort. The recording can, for example, take place with a time delay or “offline”. The recording can take place approximately once a day from smartphones that have application software specific to the application of the invention, or from connected vehicles, for example via so-called Car2X communication. A collection of sensor data for c) can in particular be carried out at least partially via suitable mobile radio communication, e.g. according to 5G standards, so that data from as many participants or vehicles as possible can be collected. Sensors used in pulse data acquisition include, for example, gyroscopes, accelerometers, magnetometers, GPS, as well as camera sensors for corresponding evaluations. The sensors used can, for example, be part of the vehicle equipment or be part of a smartphone carried.

The scope of the invention also includes recording the sensor data via external computers or servers, such as those from vehicle manufacturers that collect data from their own fleet via Car2X communication (see e.g DE 10 2020 108 531 A1 or DE 10 2019 203 405 A1 ). Neither direct communication between the computer system and the devices that generate the sensor data is required, nor real-time communication. This allows a significant reduction in communication effort and, among other things, anonymization in the sense of data protection.

Sensor data can be recorded in a filtered manner, for example only in connection with events recognized as critical by the respective device software, or more comprehensively to achieve the most comprehensive possible database, for example for AI-supported pattern recognition. Sensor data refers in particular to data Acceleration measurements that allow conclusions to be drawn about critical events or maneuvers in road traffic. The sensor data for c) can be recorded by the computer system directly or indirectly, for example using known LoT techniques or interfaces. For acquisition, the sensor data or signals are recorded and digitized using sensors, in particular acceleration sensors, whereby this is done by external devices that are not part of the computer system.

According to a special feature, sensor data for c) is recorded from different types of road users, in particular from motor vehicles (cars, trucks, etc., possibly motorcycles) and at least also from two-wheelers, in particular bicycles and also, for example, e-bikes, e-scooters and the like. , which typically form a more vulnerable type of road user. Sensor data of a first type, namely from sensor values that are generated by vehicle equipment, and sensor data of a second type are preferably recorded, which are generated in particular by portable mobile devices, preferably smartphones with appropriate software, so that sensors from road users without vehicle equipment (e.g. bicycles, E -Bikes, e-scooters etc.) is used. The acquisition of data about different types of road users is particularly advantageous, in particular for the most comprehensive detection possible, in particular early detection of danger spots, as well as for the validation of data of a different type, e.g. the validation of sensor data of a first type by sensor data of a second type.

Based on the three data categories, the method or system assigns accident data, sensor data and user input data to segments of the map-like image. This is done in accordance with the georeferencing associated with these data, such as GPS coordinates or comparable position information, in particular by the computer system.

As a result, affected segments of the map-like image have associated or segment-related recognition data, which, according to the invention, is used for early detection and/or evaluation of danger spots. The segment-related detection data includes data from at least one or more of the above three data categories a) to c), i.e. each includes accident data, sensor data and/or user input data.

As a result, a large number of the segments have at least assigned sensor data and also user input data or accident data. This creates, among other things, a A cheaper amount of training data is provided, which enables a more advantageous or more targeted application of AI-supported data processing, in particular machine learning and / or pattern recognition, or a more successful blending of the data using other information technologies. Already in the combination of the three data categories a) to c) for further processing, in particular as training data for machine learning, an independently inventive or inventive feature is seen. Partial registrations, among other things, remain expressly reserved.

Any suitable hardware architecture can be considered for the computer system used, a server system, a large-scale computer system set up for AI technologies and, in particular, a cloud architecture or, for example, computer clusters.

According to a first aspect, an evaluation can then be carried out on the basis of segment-related recognition data and at least one frequency determination and/or a data comparison can be carried out, in particular, for example, using pattern recognition.

Segments to which accident data are assigned make it easy to identify danger spots. However, a decisive advantage of the invention lies in the early detection of danger spots for which there is little or no accident data.

1. i. diesem Segment mit vorbestimmter Häufigkeit Nutzereingabedaten und Sensordaten zugeordnet worden sind;
   und/oder
2. ii. zu diesem Segment der Datenabgleich von Nutzereingabedaten und/oder Sensordaten einen vordefinierten Grad an Merkmalsübereinstimmung bzw. Merkmalskorrelation aufzeigt.

Based on the evaluation of segment-related detection data, the computer system can still identify a geo-referenced segment with detection data that does not contain accident data as a potential danger spot for the purpose of early detection, namely whether the method or system determines whether

1. i. User input data and sensor data have been assigned to this segment with a predetermined frequency;
   and or
2. ii. for this segment, the data comparison of user input data and/or sensor data shows a predefined degree of feature agreement or feature correlation.

Even a simple frequency determination can enable early detection of danger spots, for example if a predetermined, for example empirically determined, first partial frequency of sensor data that indicates critical events is combined with a second partial frequency of user input data, in particular user input data that matches the content, is present. A decision based on a total frequency of user input data and sensor data is also within the scope of the invention.

In a further, advantageous approach, a potential danger point is alternatively/or additionally concluded if a data comparison of user input data and/or sensor data, in particular the sensor data, shows a sufficient feature match or feature correlation, in particular with segments that have already been recognized as critical, for example by Accident data or user input is validated. Different approaches can be considered for suitable data comparison, in particular AI-based data processing, e.g. machine learning and/or pattern recognition, e.g. using dynamic time warping, support vector machines, random forests and/or artificial neural networks (ANN).

When comparing data, a matching criterion can in particular be determined as to whether there is a match, in particular with regard to categories recognized by pattern recognition, of information from sensor data from sources of a first type and from sensor data from sources of a second type, e.g. different types of road users. Sensor data of a first type can in particular be those from motor vehicle sensors and sensor data of a second type can in particular be from mobile devices, for example from cyclists. Based on this, a corresponding relevance information can also be assigned to the danger point under consideration. The appropriate use and evaluation of the sensor data from different types of road users is already seen as an independently inventive aspect.

Additionally or alternatively, the correspondence between information from sensor data and user input data can also be checked.

In a further step, the computer system marks all segments identified as danger spots, including those for which no accident data is available, in the map-like image.

The segments identified as danger spots are preferably assigned a computationally determined danger score, which allows users to display an indication of the level of danger. This can also be done specifically for different types of road users.

The data of the three data categories a) to c) are preferably georeferenced data that can be assigned to a (map) segment, preferably uniquely assigned. Thus, both accident data, user input data and sensor data are preferably each georeferenced. Furthermore, the data of the three data categories a) to c) are preferably qualified or assigned or assignable in a participant type-specific manner.

According to a second independent aspect, a computer-aided method or system for determining a danger score of a georeferenced, structural danger point in a digital traffic route network mapping is proposed. This can be based in particular on the system or method described above.

According to the second aspect, data is collected in particular from geo-referenced user input data related to potential danger spots, for example as explained above.

Furthermore, the aforementioned three data categories a) to c) are used and the georeferenced accident data provided, the georeferenced sensor data collected and the georeferenced user input data collected are assigned to segments, corresponding to the georeferencing associated with these data. This can be realized in particular by the computer system. As a result, segments of the map-like image are assigned segment-related recognition data, each of which includes accident data, sensor data and/or user input data.

According to the second aspect, a danger score is computationally determined for segments of the map-like image with associated recognition data by evaluating the recognition data associated with the segment under consideration.

• dass der Gefahrenscore eines georeferenzierten Segments mit Erkennungsdaten, welchem keine Unfalldaten zugeordnet sind, in Abhängigkeit von zugeordneten Nutzereingabedaten ermittelt, insbesondere validiert bzw. relativ erhöht wird, etwa im Vergleich zu Segmenten ohne zugeordnete Nutzereingabedaten;

und/oder

• dass der Gefahrenscore in Abhängigkeit einer Gewichtung der Erkennungsdaten ermittelt wird mit jeweils unterschiedlichem Gewicht für Unfalldaten, Sensordaten und Nutzereingabedaten, insbesondere mit höherem Gewicht von Unfalldaten gegenüber Nutzereingabedaten und/oder höherem Gewicht von Nutzereingabedaten gegenüber Sensordaten.

In particular, it is provided that:

• that the danger score of a georeferenced segment with detection data to which no accident data is assigned is determined, in particular validated or relatively increased, depending on assigned user input data, for example in comparison to segments without assigned user input data;

and or

• that the danger score is determined depending on a weighting of the detection data, each with a different weight for accident data, sensor data and user input data, in particular with a higher weight of accident data compared to user input data and / or a higher weight of user input data compared to sensor data.

The risk score determined in this way is then assigned or assigned to the corresponding segment. By using the above three data categories a) to c) This makes objectification and greater reliability possible when assessing the level of danger, with the advantage being that the method or system records and takes sensor data from different road users into account, so that different danger scores can also be determined and displayed for different participants.

The computational determination of a danger score for an identified danger spot can be carried out by calculation, the calculation comprising a weighting of georeferenced assigned sensor data using predefined weighting factors, and preferably at least some georeferenced sensor data of a first type, in particular originating from a first type of road user, depending on a predefined degree of agreement with features of georeferenced assigned user input data and / or further georeferenced assigned sensor data of a second type, in particular coming from a second type of road user, are weighted.

Alternatively or additionally, the danger score for an identified danger point can be calculated by processing using a computer, in particular using a neural network and/or pattern recognition or the like by the at least one computer system. This can be done in particular by taking event-related features of the infrastructure data from the map image into account.

Advantageous further developments, which are the subject of the subclaims, but are each individually inventive or are to be viewed as independently relevant to the invention, are discussed below. These are generally applicable to the two independent aspects mentioned above, and partial registration is reserved.

In an advantageous embodiment, data on different types of road users are obtained, in particular sensor data from different types of road users are recorded.

For this purpose, it is advantageous if the movement-related sensor data include a type feature which indicates which type of road user the sensor data comes from and enables a distinction between at least two different types of road users, in particular motor vehicles and two-wheelers.

Particularly but not exclusively in combination with movement-related sensor data that have a type feature, an advantageous development provides that at least for segments with recognition data for this recognition data, in particular for sensor data with type characteristics, a computer pattern recognition, in particular AI-supported pattern recognition, is carried out and a classification into critical events is carried out, in particular by the computer system, and the data comparison of user input data and / or sensor data with corresponding data to classify critical events takes place.

To collect safety-relevant events, acceleration data in particular is recorded as sensor data. This can be used to determine the respective trajectory of road users during associated critical driving maneuvers. To identify potentially critical maneuvers, for example, acceleration limit values can be set. The sensor data can accordingly be recorded using a smartphone with a suitable smartphone app, a telematics tag, or using IoT devices.

On the one hand, this data can be recorded for different types of road users and, on the other hand, it allows critical situations to be classified, located and examined for causes. This is based on data from the actual perspective of a road user, e.g. from the vehicle or two-wheeler.

To classify critical maneuvers of different road users, calculation methods from pattern recognition and machine learning are used. Various methods can be used for this, such as dynamic time warping, support vector machines, random forests and neural networks. Such methods initially make data from different types of road users comparable, which noticeably simplifies analyzes with regard to safety assessment. Deep learning approaches can also be considered, but are not mandatory, since the data from sources a) to c) are typically already obtained in a structured form or can be structured through (partially) automatic preprocessing.

In order to derive conflicts between road users that indicate danger spots, critical events identified by all road users are overlaid on a point under consideration, for example a segment. Based on these findings, a map-based forecast model can be created to estimate critical conflicts, particularly by taking into account infrastructure data and other contextual data. The degree of criticality can then be determined by comparing it with officially collected traffic accident data (see data category a) above)) as well as with citizens reported dangerous situations (see data category b) above).

A preferred embodiment therefore provides that at least for segments with recognition data, a computer pattern recognition, in particular AI-supported pattern recognition, is carried out for these recognition data, in particular for sensor data with a type feature, and a classification into critical events is carried out, in particular by the computer system. Based on this, the data comparison of user input data and/or sensor data can be carried out with corresponding data to classify critical events.

According to one aspect, the computer system can use a machine learning model obtained by AI from the three main data sources a) to c) in order to improve the map-like representation, in particular to expand the traffic network-related coverage of early detection. The hazard map can therefore provide the training data for a model, which is then used for recursive self-optimization of the hazard map. The expansion can relate to those positions or segments for which there is insufficient data from sources a) to c) or data with insufficient significance. This means that potential danger spots can be identified by comparing patterns based on infrastructure information alone. In particular, if only sensor data c) is available in conjunction with infrastructure data, early detection can be made possible by comparing patterns with validated danger spots. When creating the machine learning model, the accident data a) or user input data b) can be used in particular to validate patterns classified as critical or to optimize the model.

According to one aspect, for the purpose of early detection, the computer system can identify a georeferenced segment with detection data to which no accident data is assigned as a potential danger spot if a data comparison of the sensor data for this segment shows sufficient feature agreement or feature correlation with sensor data of a critical event, for determining the feature match or feature correlation, the upstream pattern recognition is used.

Regardless of the type of data comparison used, it can be provided in a data-technically favorable implementation that a data object is assigned to segments identified as danger points, which has at least one includes a computationally determined risk score.

The specific programming implementation of the software, in particular data objects, is not crucial, but an object-oriented data structure or object-oriented programming is particularly preferred.

Particularly preferably, each data object assigned to a danger point has several participant type-specific danger scores, so that danger points are evaluated or can be evaluated from the perspective of different participants in the sense of a holistic approach.

• zumindest eine Relevanzangabe dahingehend für welchen bzw. welche von mehreren verschiedenen Verkehrsteilnehmertypen, insbesondere Fußgänger, Radfahrer oder Kraftfahrzeugführer, eine Gefahr droht, wobei die Relevanzangabe vorzugsweise aus den zugeordneten Erkennungsdaten bestimmt wird; und/oder
• zumindest eine Gefahrenartangabe dahingehend, bei welchem Verkehrsverhalten, welcher Situation und/oder zwischen welchen Verkehrsteilnehmern an der Gefahrenstelle Gefahr droht, wobei die Gefahrenartangabe vorzugsweise aus den zugeordneten Erkennungsdaten bestimmt wird; denn vorzugsweise geben die Erkennungsdaten neben der Gefahrenart ebenfalls Aufschluss hinsichtlich der spezifischen Gefahrensituation und ggf. der Konflikte zwischen Verkehrsteilnehmern und/oder
• zumindest eine Kontextangabe dahingehend, bei welchen Umwelteinflüssen bzw. Verkehrssituationen, an der Gefahrenstelle Gefahr droht, wobei die Kontextangabe vorzugsweise bestimmt wird aus den zugeordneten Erkennungsdaten und/oder aus zusätzlichen Verkehrswegmetadaten, insbesondere umfassend Information bezüglich Wetterabhängigkeiten, Geschwindigkeitsrichtwertem, zeitabhängigem Verkehrsaufkommen oder dgl..

It can be provided that the data object further includes:

• at least one relevance indication as to which of several different types of road users, in particular pedestrians, cyclists or motor vehicle drivers, are at risk of danger, the relevance indication preferably being determined from the associated recognition data; and or
• at least one danger type indication as to which traffic behavior, which situation and/or between which road users at the danger zone there is danger, the danger type indication preferably being determined from the associated recognition data; Because preferably, in addition to the type of danger, the detection data also provides information regarding the specific dangerous situation and, if necessary, the conflicts between road users and/or
• at least one contextual indication as to which environmental influences or traffic situations pose a threat of danger at the danger point, the contextual indication preferably being determined from the assigned recognition data and/or from additional traffic route metadata, in particular comprising information regarding weather dependencies, speed guidelines, time-dependent traffic volume or the like.

In a preferred further development to achieve a holistic view or a system that is useful for different types of participants, it is provided that the computational determination of a danger score is specific to the participant type or depending on the type of road user and each determined risk score is assigned to exactly one of several predefined types of road user. This allows the traffic situation to be viewed holistically from the different perspectives of all road users. This is not just about motor vehicle drivers (currently the focus of many assistance systems), but especially about them Weaker and therefore more vulnerable road users such as pedestrians, cyclists, Pedelec riders, e-scooter riders, etc. However, participants who are less physically at risk are also considered, for example truck, bus or tram drivers - for whom an accident can have significant psychological consequences.

The computational determination of a danger score in a participant type-specific manner can create a participant type-specific, multidimensional space for each segment or data object, in which the various danger information can be aggregated from all perspectives, evaluated and also made individually retrievable as a danger score. For this purpose, it is advantageous if the danger scores are linked to the segment or data object in a mathematical sense in a multidimensional manner on participant type-specific axes. A computationally simple implementation can be achieved, for example, if the dimensions of a vector space (including Hamel dimensions) are assigned to the participant types, so that effective, common matrix calculation can be used to determine the danger scores. The vector space defined in this way for each segment or data object then has at least one Hamel dimension for each type of road user to be considered.

Furthermore, within the road user types, subtypes can also be assessed differently, e.g. road user type: pedestrian, subtype: child/adult/seniors. This approach allows, for example, child-specific safe routes to school for children or senior-friendly routes (e.g. where hearing and seeing or quick crossing are less relevant) to be determined.

The system can continuously and, in particular, assess participant-specifically, i.e. mutually from different perspectives, whether there are any noticeable patterns.

It is therefore advantageous if the data objects for danger spots at least indicate which road users are in danger or are at risk (relevance information) and what exactly the danger is (indication of the type of danger). Any type of data set can be considered for storing the data or as data objects, whereby the data from the different sources a) to c) can, if necessary, be brought into a standardized format through preprocessing, for example in the course of AI-supported pattern recognition or the like .

With regard to data communication, one embodiment provides that Data collection of georeferenced user input data related to potential danger spots is carried out continuously and/or sporadically via an Internet interface of the computer system. This Internet interface can be connected to at least one or more servers, in particular web server(s), each of which provides a user interface for user input data. In this way, several possibly different interfaces can be provided, for example a classic website and an interface for smartphone apps. It is also possible, for example, to integrate insurance accident reports from road users via a corresponding interface.

According to a further advantageous aspect, the map-like image is divided into a large number of georeferenced segments in a variable and possibly iterative manner. Variable subdivision into segments of different geographical lengths can therefore be provided, in particular depending on the traffic route infrastructure (e.g. the type of route with longer segments, e.g. on motorways) and/or depending on assigned detection data and/or depending on identified danger spots. This enables, among other things, sufficient local discrimination of danger spots.

Furthermore, assigning accident data, sensor data and user input data to individual segments can include determining a geographical distance between the detection data from their georeferencing, so that the assignment can be made based on relative distances. This can also, among other things, improve local discrimination of danger spots.

Since recording data from different road users is seen as particularly advantageous, one embodiment for data acquisition provides that sensor data is recorded based on acceleration measurement values from different sources, in particular sensor data of a first type, which consists of sensor values from the vehicle equipment, in particular vehicle sensor systems or telematics equipment. are generated by vehicles, in particular motor vehicles, and at least sensor data of a second type, which are generated from sensor values from acceleration sensors in portable mobile devices carried by road users, in particular smartphones. Other sources are also possible.

The sensor data is preferably provided with time stamps, among other things. The timestamps can be used as a feature in data comparison with regard to a predefined degree of feature agreement or feature correlation, in particular whether the timestamps are spaced apart within a predefined time range or correlate based on the time of day.

• die Smartphone-Applikation vorzugsweise eingerichtet ist, potentiell sicherheitskritische Ereignisse zu erkennen, insbesondere auf Grundlage einer Überschreitung von Beschleunigungsschwellwerten, und georeferenzierte Sensordaten lediglich zu potentiell sicherheitskritischen Ereignissen bereitstellt, wobei die bereitstellten Sensordaten vorzugsweise Informationen zum Bewegungsverhalten innerhalb eines Zeitfensters um das sicherheitskritische Ereignis enthalten; und/oder
• die Smartphone-Applikation eine Schnittstelle für georeferenzierte Nutzereingabedaten bezogen auf potentielle Gefahrenstellen bereitstellt.

According to a further, possibly independently advantageous aspect, the invention provides that georeferenced sensor data is provided by an application-specific smartphone application for detecting potential danger spots, whereby

• the smartphone application is preferably set up to detect potentially safety-critical events, in particular on the basis of exceeding acceleration threshold values, and provides geo-referenced sensor data only for potentially safety-critical events, the sensor data provided preferably containing information about movement behavior within a time window around the safety-critical event; and or
• the smartphone application provides an interface for georeferenced user input data related to potential danger spots.

• die Smartphone-Applikation bei Erkennung eines potentiell sicherheitskritischen Ereignisses entsprechenden Sensordaten zugehörige GPS-Koordinaten zur Georeferenzierung, sowie vorzugsweise zumindest einen Zeitstempel zuordnet und mit den Sensordaten dem Rechnersystem übermittelt;
• die Smartphone-Applikation, insbesondere bei Erkennung eines potentiell sicherheitskritischen Ereignisses und/oder bei Erkennung örtlicher Nähe zu einem als potentielle Gefahrenstelle identifizierten Segments, den Nutzer zur Eingabe von Nutzereingabedaten auffordert, umfassend zumindest eine Bestätigung hinsichtlich des Bestehens einer Gefahrenstelle, vorzugsweise weiterhin zur Eingabe einer Relevanzangabe bzgl. betroffener Verkehrsteilnehmertypen und/oder einer Gefahrenartangabe und/oder einer Kontextangabe und die Nutzereingaben als Nutzereingabedaten mit den Sensordaten dem Rechnersystem übermittelt; und/oder
• die Smartphone-Applikation auf Grundlage eines mittels integrierten Beschleunigungssensor ermittelten Bewegungsprofils zu einem potentiell sicherheitskritischen Ereignis eine Relevanzangabe bzgl. betroffener Verkehrsteilnehmertypen bereitstellt und mit den Sensordaten dem Rechnersystem übermittelt.

When using a smartphone application specific to the application of the invention to detect potential danger spots, it can also be advantageously provided that

• When a potentially safety-critical event is detected, the smartphone application assigns corresponding sensor data to the associated GPS coordinates for georeferencing, and preferably at least one time stamp, and transmits it to the computer system with the sensor data;
• the smartphone application, in particular when detecting a potentially safety-critical event and / or when detecting local proximity to a segment identified as a potential danger point, asks the user to enter user input data, comprising at least one confirmation regarding the existence of a danger point, preferably further to enter a Relevance information regarding affected road user types and/or a hazard type information and/or a context information and the user input is transmitted to the computer system as user input data with the sensor data; and or
• the smartphone application, based on a movement profile determined by means of an integrated acceleration sensor, provides relevance information regarding affected road user types for a potentially safety-critical event and transmits it to the computer system with the sensor data.

The information preferably assigned to each identified danger point preferably includes information about the type of danger. Here, a computational determination of at least one hazard type information can be provided, the hazard type information being determined from information from infrastructure data and/or sensor data and/or accident reports and/or user input data assigned to the segment, preferably from a combination of information from at least two different of these data.

If necessary, danger scores calculated with higher resolution can be represented more ergonomically by dividing them into a reduced number of danger levels. It is therefore possible to assign a danger level from a number of discrete danger levels to each danger score, with each danger level being assigned a predefined value range of the danger score.

Furthermore, a number of danger scores can be determined computationally for an identified danger point depending on the respective relevance information as to which of several different types of road users, in particular pedestrians, cyclists or motor vehicle drivers, are at risk. This means that user-specific danger scores can be assigned to different types of road users.

In an advantageous implementation, the data included in the digital mapping of traffic route networks, in particular the danger scores, is gradually updated when a predetermined amount of further sensor data has been recorded and/or a predetermined amount of further user input data has been collected. This means that the danger zone map can be optimized step by step and continuously updated without requiring significant computing effort, for example for new pattern recognition.

In a variant, it can be provided that the data collection of geo-referenced sensor data, based on the movement behavior of a road user, takes place in a different manner when the local proximity of the road user to a segment that has already been identified as a potential danger spot is detected, so the data collection of sensor data can then be carried out in particular with increased Sensitivity or intensity. This can be achieved, for example, through suitable programming of a smartphone application and enables improved information on potential danger spots, which in turn enables positive feedback to improve pattern recognition.

According to a further independent aspect, the invention relates to a digital hazard map that was generated according to one of the embodiments discussed above, i.e. with segments identified as danger spots, in particular danger spots identified by early detection for which no accident data is available, and / or segments with a danger score, in particular based on Danger scores determined from the three data sources a) to c).

Accordingly, the invention also relates to the use of the resulting digital hazard map (map-like digital traffic route network mapping), which has segments of the traffic routes identified as danger spots according to one of the above approaches using georeferenced accident data provided by, collected georeferenced user input data and collected georeferenced sensor data.

The invention therefore also relates in particular to a computer system, in particular a cloud computer system, comprising a map-like digital traffic route network image, which was generated according to a method according to one of the embodiments discussed above, and in particular has segments with a danger score that are identified as danger spots.

This system can be used in particular to generate hazard-related warning messages to road users. For this purpose, terminal devices communicating with the computer system, including terminal devices installed in vehicles (e.g. navigation devices), but in particular smartphones, can issue a warning message when local proximity to a segment identified as a danger point is detected, in particular a warning message comprising an indication of relevance and/or hazard type information and/or context information. By sending appropriate warnings to road users, especially when danger spots are detected early, road safety can be noticeably improved overall, especially for types of road users indirectly affected by the warning, e.g. pedestrians or cyclists. The output of warning messages can in particular be specific to the participant type or depending on the type of road user, and for this purpose, for example, use the relevance information or the like.

The resulting hazard map can also be used advantageously in various other applications.

An advantageous application lies, for example, in determining safe route options in a navigation system, in particular in addition to fast and/or ecological route options determined by the navigation system. This can then also take into account previously identified danger spots. The use of such a navigation system is not only advantageous for driving road users, but also for non-motorized road users, e.g. for determining safe routes to school for children. This can also be done “offline” via a web-based or server solution, for example.

Another advantageous application lies in the optimization of solutions for autonomous or semi-autonomous driving. The system can be used to communicate, in particular information relevant to danger zones from the digital hazard map, with autonomous or semi-autonomous driving vehicles and, on this basis, allows targeted influencing of the autonomous or semi-autonomous driving behavior depending on danger zones, in particular taking into account the relevance and/or type of danger - and/or context information and/or to influence route selection. In this way, driving behavior can be adapted automatically, specifically and better to areas with high risk potential.

Here, data that was generated according to the invention, for example data from the digital hazard map or danger spots identified and/or evaluated with the system, in particular as Part of the "Operational Design Domain" (ODD: in the sense of SAE INTERNATIONAL. Taxonomy and definitions for terms related to driving automation systems for on-road motor vehicles. SAE international, 2018, 4970. Vol., No. 724, p. 1 -5 ) be used.

Here, data that were generated according to the invention can also be used particularly advantageously to determine and/or optimize the ODD of an autonomously driving vehicle (AV), in particular an autonomously driving vehicle of the L3+ type, in particular an unlimited ODD of an L5-AV (cf. SAE ibid). become.

Another advantageous application lies in the provision of training data sets for training a machine learning model to predict potential danger spots, e.g. for recursive optimization or expansion of the coverage in the digital danger map itself.

The resulting machine learning model can also be used in particular in a traffic planning simulation. This allows improved analysis tools for Traffic planners or authorities can implement hazard simulation during planning or can significantly improve it.

Finally, the invention also relates to a computer system, in particular a cloud computer system, comprising an arrangement with at least one processor and at least one memory, the arrangement being set up or configured according to the invention to carry out a method, with at least the features of claim 1 or Claim 2 and possibly also according to one of the dependent claims.

BRIEF DESCRIPTION OF THE FIGURES

• FIG. 1 ein schematisches Grundkonzept zur Datenverarbeitung, insbesondere zur Bestimmung eines sog. Gefahrenscore für ein Verkehrswegenetz;
• FIG. 2 zeigt ein Datenflussschema zur datentechnischen Veranschaulichung einer Ausführungsform des erfindungsgemäßen Verfahrens zur Früherkennung von Gefahrenstellen sowie des erfindungsgemäßen Rechnersystems;
• FIG. 3 zeigt ein Flussdiagramm bzw. Ablaufdiagramm einer Ausführungsform des erfindungsgemäßen Verfahrens zur Früherkennung und Bewertung von Gefahrenstellen;
• FIG. 4 zeigt rein beispielhafte User-Interface Ansichten einer App bzw. Webseite für gefahrenstellenbezogene Nutzereingaben die dem Rechnersystem bzw. Verfahren zugeführt bzw. von diesen erfasst werden; und
• FIG. 5 zeigt eine beispielhafte Darstellung einer Gefahrenstellenkarte mit anhand von diskreten Gefahrenlevels angezeigten Gefahrenstellen.

Further features and advantages of the invention can be found in the following description of a preferred exemplary embodiment based on the accompanying figures, without limiting the scope of protection. Show here:

• FIG. 1 a basic schematic concept for data processing, in particular for determining a so-called danger score for a traffic route network;
• FIG. 2 shows a data flow diagram for the data technical illustration of an embodiment of the method according to the invention for the early detection of danger spots and of the computer system according to the invention;
• FIG. 3 shows a flow chart or flowchart of an embodiment of the method according to the invention for early detection and evaluation of danger spots;
• FIG. 4 shows purely exemplary user interface views of an app or website for hazard-related user input that is fed to or recorded by the computer system or process; and
• FIG. 5 shows an exemplary representation of a danger zone map with danger zones displayed based on discrete danger levels.

DETAILED DESCRIPTION BY THE FIGURES

FIG. 1 outlines the underlying concept of an embodiment of the invention. This embodiment provides, as a core feature, the processing and analysis of data from three independent and different data sources for a traffic route network-wide hazard point identification and/or hazard score calculation. The first data source is accident data 1, in particular from the police and/or from insurance companies Registered accident data is used, which is recorded, for example, by the police authorities via IT systems. This at least covers more serious traffic accidents that have already occurred, which can also affect various road users. From such accident data 1 alone, no statement can be made about dangerous areas where no accident has yet occurred. Furthermore, in the case of minor accidents, the police are often not informed, so that these accident sites do not appear in the statistics, or the recording of such accidents by authorities is not required.

The resulting unreported number of minor accidents as well as near misses and other conflict situations should be uncovered via two additional data sources, namely on the one hand via danger reports from road users in the form of user input data 3 and via sensor data 2 (so-called pulse data), which come from motor vehicles with suitable sensor equipment and/ or from smartphones, especially during critical driving maneuvers (e.g. swerving or braking sharply). A web platform and/or a smartphone app is provided to generate danger reports from user input data 3 (cf. FIG.4 ). Here, road users can report danger spots for the entire transport network on a digital representation of transport route networks in the form of a digital, interactive map (see FIG. 4 ), comment on and validate, falsify and/or evaluate already recorded danger spots. This means that danger spots can be identified throughout the transport network through “crowdsourcing”.

FIG. 2 shows a basic data flow diagram, based on which the process of an embodiment of a method according to the invention for identifying a danger point in a digital image of traffic route networks, or the so-called danger map, is explained. The approach is based on the in FIG. 1 presented concept. Four data sources are shown schematically here in the form of digitized georeferenced accident data 1, georeferenced sensor data 2, which sensors generate that are carried by road users or vehicles, georeferenced user input data 3 related to potential danger spots, as well as optional, georeferenced context data 4. Data from these sources 1, 2, 3, 4 are from . processed by at least one computer system, for example a cloud computer system 5. In the present case, georeferenced means that at least one piece of information related to the geographical position is assigned to the data, for example GPS coordinates or the like.

The digitized georeferenced accident data 1 is provided to the cloud computer system 5, for example in the form of a database 1. This georeferenced accident data 1 may have been collected or recorded in advance by authorities, such as road traffic offices and/or police authorities, and/or, for example, by insurance companies and/or vehicle rental companies and made available by them; if necessary, data is processed in a suitable manner for this purpose unified formats.

The geo-referenced sensor data 2 are continuously and/or sporadically recorded directly or indirectly by the cloud computer system 5, with these geo-referenced sensor data 2 from the electronics of vehicles participating in traffic and/or preferably from devices which road users, such as drivers, Cyclists or e-scooter drivers carry with them, in particular smartphones, smartwatches, telematics tags, navigation systems, IoT devices, etc. generated. The sensor data 2 (e.g. from a gyroscope, accelerometer, magnetometer and camera sensors) include movement values, in particular acceleration values, and position information data assigned to them, e.g. GPS coordinates. The sensor data 2 can be recorded, for example, by one or more servers, not shown, with which the cloud computer system 5 communicates. Furthermore, the sensor data 2 can preferably include time information or time stamps, which indicate the time of data generation. In an advantageous embodiment, the sensor data 2 can also include information about the corresponding type of road user, which results inherently from the data source, for example in a vehicle navigation system or a telematics system, or can be supplemented by the same, for example a smartphone APP. Georeferenced sensor data 2 can be continuously, automatically collected by the system.

The georeferenced user input data 3 related to potential danger spots is collected user-initiated or system-initiated, with or without a direct time commitment. The user input data 3 can also be collected via an intermediate server (not shown), for example a web server, with which the cloud computer system 5 communicates. The input can be user input data 3 can be initiated by a user. This user input data 3 includes subjective information about at least one potential danger point perceived as such by the user, the information comprising georeferenced position information, for example GPS coordinates of the potential danger point and/or information regarding the type of danger of the potential danger point and/or information regarding a Relevance of the potential danger spot for one or more different types of road users.

Optional georeferenced context data 4 are automatically recorded continuously and/or sporadically, with the context data 4 comprising temporary information, in particular time-dependent information, such as weather data and/or information about the traffic situation and/or information about the season.

The cloud computer system 5 processes the recorded georeferenced data to determine a danger point based on the georeferenced position information assigned to the data. If a danger point has been identified, the cloud computer system 5 creates or changes a corresponding data record or a corresponding data object in the danger map.

Furthermore, the cloud computer system 5 can trigger geo-referenced warning messages or warnings 7 related to a danger point, which are received, interpreted and, in particular, output visually and/or acoustically to a user, for example visually and/or acoustically, by terminal devices 6, in particular when they are geographically approaching this danger point can be. The information from the cloud computer system 5 can also be used to communicate with autonomous or semi-autonomous driving vehicles in order to influence the autonomous or semi-autonomous driving behavior depending on danger spots. Optionally, the warning 7 can trigger a user query, which causes terminal devices 6 to ask the user for verification, falsification and/or a comment related to the danger point. The user response 8 to such a user query can be transmitted from terminal devices 6 to the cloud computer system 5 for verification, falsification and/or evaluation of the danger spots. So user input data 3 and/or user responses 8 can be processed by the cloud computer system 5, for example to check danger spots that have already been identified by sensor data 2 and/or to modify danger spots that have already been identified.

According to one implementation, it is advantageously provided that an adapted type of output of the information is also made according to the distinction according to type of road user: the relevant information, for example, can be transmitted to the road users acoustically, visually and/or haptically. Depending on the road user, a wide variety of actuator units come into consideration, for example smart watches, smartphones, steering wheel grips, warning sounders (bell/horn, etc.), lighting on the outside of the vehicle, interior lighting in the vehicle, etc. Particularly preferred is an automatic setting of the output type and the type of road user Output actuator (e.g. acoustically and haptically to pedestrians and cyclists via Smartphone), visual and broadcast. The warning can also be directed at other road users and does not have to be aimed at the system user themselves (e.g. automatic bicycle bell when crossing a pedestrian crossing).

FIG. 3 shows a flowchart of an embodiment of a method according to the invention for identifying and verifying a danger spot using georeferenced information from several different data sources, in particular accident data 1 and sensor data 2 and user input data 3, each if available.

In the process after FIG.3 , in a first step 301, software assigns information from georeferenced detection data, accident data 1 and/or sensor data 2 and/or user input data 3, corresponding to the georeferenced position information, to a segment of the digital map.

After the information has been assigned, a comparison of the information assigned to a georeferenced potential danger point follows in a second step 302, in particular a comparison of information from at least two different data sources 1, 2, 3. If a predefined degree of agreement of the information assigned to a potential danger point, e.g. from at least two of the three different data sources 1, 2, 3, and/or sufficient partial frequencies of critical sensor data 2 and/or user input data 3 are reached and/or if information from accident data 1 is available, then a third step 303 can be continued. Since information from accident data 1 can be viewed as officially validated, it generally does not require any verification and can be used as a basis for marking or identifying (already known) danger spots.

In the third step 303, a segment of the digital danger map is identified as a danger point and provided with corresponding data in the data record or data object, including a relevance indication as to which type of road user is at risk of danger, a danger type indication, for example as to which traffic behavior poses a danger, and a contextual indication to that effect , under which environmental influences or traffic situations there is a risk of danger at the danger point. To obtain this data, the recognition data 1, 2, 3 and/or additional traffic route metadata, in particular comprehensive information regarding weather dependencies, speed guidelines, time-dependent traffic volume, etc. can be used.

If the predefined degree of agreement is not achieved in the second step 302 and there is no relevant accident data 1, a further test step 303 'follows. If, for example, the information situation from sensor data 2 or user input data 3 for a segment to which no accident data 1 is assigned is not sufficiently clear, a special data comparison can take place in step 303 '.

In step 303', in particular, user input data 2 and/or sensor data 3 are examined for a predefined degree of feature agreement or feature correlation. For the purpose of early detection, the computer system 5 can identify a georeferenced segment without accident data 1 as a potential danger spot if a data comparison of the sensor data 3 for this segment shows sufficient feature agreement or feature correlation with sensor data 3 of a critical event, which is checked in a further test step 304 '.

As a basis for the computational check for feature agreement or feature correlation, the computer system 5 can, for example, use previously carried out AI-based pattern recognition, on the basis of which all recorded geo-referenced recognition data 1, 2, 3 are classified into critical events. A technique known per se can be used for pattern recognition, e.g. pattern recognition using dynamic time warping, support vector machines, random forests and/or artificial neural networks (ANN).

In step 303', based on a classification obtained in this way, the data comparison of sensor data 3 can be carried out with corresponding data to classify critical events.

If the vote in test step 304 'shows that there is sufficient feature agreement or feature correlation, the segment is again identified as a danger point in the third step 303 and provided with corresponding data. If in step 304' no sufficient feature match or feature correlation is found, this can be used, for example, for model optimization in step 305'.

In the third step 303, additional information, for example from georeferenced context data 4, can be assigned to the georeferenced potential danger point. Additionally or alternatively, in the third step 303, a user request is generated, which asks at least one user to validate the potential danger point by entering 4.

In a fourth step 304, the data on the segment is saved and an evaluation is carried out by calculating a danger score. For this purpose, the danger score can be calculated by determining it depending on a weighting of the detection data, each with a different weight for accident data 1, sensor data 2 and user input data 3, in particular with a higher weight of accident data 1 compared to user input data 3 and / or a higher weight of user input data 3 compared to Sensor data 1.

The hazard map can then be updated in a fifth step 305.

An important data source for the method according to the invention is user input data 3. Experiments have shown that the so-called crowdsourcing approach can in principle be used for proactive detection of danger spots in road traffic. On an interactive danger zone map, such as in FIG. 4 Presented in the form of an app and a homepage, users can submit danger reports.

FIG. 4 shows an embodiment of a digital hazard map in an app or a homepage for generating user input data 3, which can be used in a method according to the invention (see FIG. 2 Reference number 3). The user has the option of providing information in the form of a hazard relevance 401 (relevance information), a hazard type 402 (danger type information) and the exact geographical position of the danger point 405 (georeferencing). The danger relevance 401 indicates which types of road users the danger spot represents a danger for. Furthermore, the danger zone map offers the possibility of storing comment texts 403 and images 404 related to a danger zone, so that, for example, a danger zone can be described in a comprehensible manner for other users, even if they are from outside the location and/or to provide further information for any authorities/offices. This user input data 3 of the danger zone map can then be used, for example, by an embodiment of the method according to the invention FIG. 2 be collected by the cloud computer system 5. Furthermore, one embodiment of a danger zone map offers the possibility of evaluating already identified danger zones for validation and modification and/or requests a user to validate an identified danger zone (see FIG. 2 reference number 8). Marking a danger spot through user input on the interactive map, for example, takes place in three main steps: First, the road user marks the corresponding spot on the map and can then determine the type of danger, for example whether the spot is confusing or has poor road conditions (402). In the second step, you can then select for whom the location poses a source of danger: e.g. pedestrians, cyclists, motorcyclists and motor vehicle drivers (car, truck or bus drivers) and thus generates a hazard relevance 401 (relevance statement). In the next step, the more precise danger trigger can be specified, such as unclear traffic routing or wild growth on the side of the road. Furthermore, the app or website (not shown) can provide the user with contextual information as to which environmental influences or traffic situations typically pose a danger at the danger point.

FIG. 5 shows an example of the danger levels 502 determined by means of a method according to the invention in a danger zone map 500 for an urban area, the determined danger levels 502 being compared to the accident clusters 501 that were determined by the accident commission for the same period of time. The danger levels are graded in color: the darker, the more dangerous a section or junction is. The accident sites or accident clusters are shown as circles 501. It can be seen that accident sites or accident clusters 501 only cover a portion of the road sections and junctions 502 identified as dangerous. In addition, the danger level 503 varies between 5 levels (from 1 to 5) at the accident clusters 501. This means that a distinction is not only made between the presence of a dangerous place and the absence of a dangerous place. In this way, the danger score can be used to determine further information about road safety and present it ergonomically for users. In addition to danger spots that exist at all times regardless of environmental factors, there are danger spots that only exist in certain contextual situations, for example during heavy rain, at night or when there is high traffic. To ensure that warnings are only issued in the appropriate context situation, the database is enriched with further context data 4 such as weather data and traffic data. In this way, the danger assessment using a danger score can dynamically take the respective context situation into account and convey the optimal information content for the respective user groups. By means of one, by way of example and excerpts in FIG. 5 The danger zone map 500 shown can be used to verify the results of the danger levels 502 determined by a method according to the invention and at the same time show that the accident data 1 in the form of accident clusters (circle) 501 are not sufficient to identify as many dangerous places as possible in a traffic route network. Tests showed that the hazard map determined using a method according to the invention is also suitable, among other things, for the early detection of structural hazards in road traffic. <b><u>Reference symbol list</u></b> FIG.1-3 304' second test step 1 Accident data 305' Optimization step 2 Sensor data FIG.4 3 User input data 401 Hazard relevance (relevance statement) 4 Context data 5 Cloud computer system 402 Danger type (danger type information) 6 End devices 403 Comment text 7 warning 404 Picture 8th User response 405 Position of the danger point (georeferencing) FIG.3 301 first step FIG.5 302 second step 500 Hazardous location map 303 Third step 501 Accident accumulation site 304 fourth step 502 Road section/junction 305 fifth step 503 Danger level 303' first test step

Claims (16)

Computer-aided method for the early detection of structural danger spots in road traffic using a digital traffic route network map using a computer system, the method comprising the following steps: - Providing a digital map-like image of at least one traffic route network, in particular comprising infrastructure data, the map-like image being loadable or being loaded into a computer memory of the computer system; - dividing the map-like image into a large number of georeferenced segments of traffic routes by the computer system; - Providing digitized geo-referenced accident data, comprising statistical accident data, in particular in the form of at least one database, which can be processed by the computer system; - Data collection, in particular by the computer system, of geo-referenced user input data related to potential danger areas; - Data acquisition, in particular by the computer system, of georeferenced sensor data related to movement characteristics of road users, in particular based on acceleration measurements that can be generated or are generated by sensors that are carried by road users or vehicles; - Assigning accident data, user input data and sensor data to segments, corresponding to georeferencing associated with these data, by the computer system, so that segments of the map-like image have associated recognition data, each comprising accident data, sensor data and / or user input data; - whereby the computer system carries out an evaluation of these recognition data at least for segments with recognition data, comprising at least one frequency determination and/or a data comparison, and whereby the computer system identifies a georeferenced segment with recognition data to which no accident data is assigned as a potential danger spot for the purpose of early detection, - if user input data and sensor data have been assigned to this segment with a predetermined frequency;
and or - if a data comparison of user input data and/or sensor data shows a predefined degree of feature agreement or feature correlation for this segment; and - Wherein the computer system identifies segments identified as danger spots in the map-like image and preferably assigns them a computationally determined danger score. Computer-aided method for determining a danger score of a georeferenced, structural danger point in a digital traffic network map, in particular a segment identified as a danger point according to the method according to claim 1, by at least one computer system, the method comprising at least: - Data collection of geo-referenced user input data related to potential danger spots; - Assigning provided georeferenced accident data, collected georeferenced user input data and collected georeferenced sensor data to segments, corresponding to georeferencing associated with these data, by the computer system, so that segments of the map-like image have associated recognition data, each comprising accident data, sensor data and / or user input data; - computationally determining a danger score for segments of the map-like image with associated detection data by evaluating this detection data, - wherein the danger score of a georeferenced segment with detection data to which no accident data is assigned is determined, in particular validated or relatively increased, as a function of assigned user input data;
and or - wherein the danger score is determined depending on a weighting of the detection data, each with a different weight for accident data, sensor data and user input data, in particular with a higher weight of accident data compared to user input data and / or a higher weight of user input data compared to sensor data; - Assigning the determined danger scores to the segments with assigned detection data. Method according to claim 1 or 2, characterized in that the movement-related sensor data include a type feature which indicates which type of road user the sensor data comes from and enables a distinction between at least two different types of road users, in particular motor vehicles and two-wheelers; and or at least for segments with recognition data for these recognition data, in particular for sensor data with type characteristics, a computer pattern recognition, in particular AI-supported pattern recognition, is carried out and a classification into critical events takes place, in particular by the computer system, and the data comparison of user input data and/or sensor data with corresponding data to classify critical events takes place, in particular, for the purpose of early detection, the computer system identifies a georeferenced segment with detection data to which no accident data is assigned as a potential danger point if a data comparison of the sensor data for this segment shows sufficient feature correspondence or feature correlation with sensor data of a critical event. Method according to claim 1, 2 or 3, characterized in that segments identified as danger spots are each assigned a data object which comprises at least one computationally determined danger score, the data object preferably further comprising: - at least one relevance indication as to which of several different types of road users, in particular pedestrians, cyclists or motor vehicle drivers, are at risk of danger, the relevance indication preferably being determined from the associated recognition data; and or - at least one hazard type information as to which traffic behavior, which situation and/or between which road users are at risk of danger at the danger point, the hazard type information preferably being from the assigned ones recognition data is determined; and or - at least one contextual indication as to which environmental influences or traffic situations pose a threat of danger at the danger point, the contextual indication preferably being determined from the assigned recognition data and/or from additional traffic route metadata, in particular comprising information regarding weather dependencies, speed guidelines, time-dependent traffic volume or the like; and or whereby each data object assigned to a danger point has several participant type-specific danger scores. Method according to one of claims 1 to 4, in particular according to claim 3 or 4, characterized in that the computational determination of a danger score is specific to the participant type or depending on the type of road user, and each determined risk score is assigned to exactly one of several predefined road user types; and or the danger scores are linked to the segment or data object multidimensionally on participant type-specific axes; and or the data collection of geo-referenced user input data related to potential danger spots is carried out continuously and/or sporadically via an Internet interface of the computer system, which is connected to at least one server, in particular a web server, which provides a user interface for user input data. Method according to one of claims 1 to 5, characterized in that
the subdivision of the map-like image into a plurality of georeferenced segments includes a variable subdivision into segments of different geographical lengths, in particular depending on traffic infrastructure and/or depending on assigned detection data and/or depending on identified danger spots, so that sufficient local discrimination of danger spots takes place; and/or assigning accident data, sensor data and user input data to individual segments includes determining a geographical distance between the detection data from their georeferencing and the assignment is based on relative distances. Method according to one of claims 1 to 6, characterized in that for data acquisition, sensor data is acquired based on acceleration measured values from different sources, in particular sensor data of a first type, which are derived from sensor values from the vehicle equipment, in particular vehicle sensors or telematics equipment, are generated by vehicles, in particular motor vehicles, and at least sensor data of a second type, which are generated from sensor values from acceleration sensors in portable mobile devices, in particular smartphones, carried by road users; and/or the sensor data are preferably provided with time stamps, the time stamps being used in the data comparison with regard to a predefined degree of feature agreement or feature correlation, in particular as to whether the time stamps are spaced apart in time within a predefined time range or correlate based on the time of day. Method according to one of claims 1 to 7, in particular according to claim 7, characterized in that
Geo-referenced sensor data are provided by an application-specific smartphone application for the detection of potential danger spots, whereby - the smartphone application is preferably set up to detect potentially safety-critical events, in particular on the basis of exceeding acceleration threshold values, and only provides georeferenced sensor data for potentially safety-critical events, the sensor data provided preferably containing information about movement behavior within a time window around the safety-critical event; and or - the smartphone application provides an interface for georeferenced user input data related to potential danger spots. Method according to claim 8, characterized in that - When a potentially safety-critical event is detected, the smartphone application assigns corresponding sensor data to the associated GPS coordinates for georeferencing, and preferably at least one time stamp, and transmits them to the computer system with the sensor data; - the smartphone application, in particular when detecting a potentially safety-critical event and / or when detecting local proximity to a segment identified as a potential danger point, asks the user to enter user input data, comprising at least one confirmation regarding the existence of a danger point, preferably further for input a relevance information regarding affected road user types and/or a hazard type information and/or a context information and the user input is transmitted to the computer system as user input data with the sensor data; and or - the smartphone application based on a movement profile determined using an integrated acceleration sensor to a potential Safety-critical event provides relevance information regarding affected road user types and transmits it to the computer system with the sensor data. Method according to one of the preceding claims, characterized by computationally determining a danger score for an identified danger point by calculating, the calculation comprising a weighting of georeferenced assigned sensor data using predefined weighting factors, and preferably at least some georeferenced sensor data of a first type depending on a predefined degree are weighted for agreement with features of georeferenced assigned user input data and/or further georeferenced assigned sensor data of a second type;
and or
computationally determining a danger score for an identified danger point by calculating, the calculation comprising processing by means of a computer, in particular using a neural network and/or pattern recognition by the at least one computer system, taking into account event-related characteristics of the infrastructure data. Method according to one of the preceding claims, in particular according to claim 4, characterized by computationally determining at least one hazard type information for each identified hazard location, the hazard type information consisting of information from infrastructure data and / or sensor data and / or accident reports and / or user input data assigned to the segment, preferably from a Combination of information from at least two different of these data is determined. Method according to one of the preceding claims, characterized by - Assigning a danger level from a number of discrete danger levels to each danger score, with each danger level being assigned a predefined value range of the danger score; and or - computationally determining several danger scores for an identified danger point depending on the respective relevance information as to which of several different types of road users, in particular pedestrians, cyclists or motor vehicle drivers, are at risk of danger and assigning a respective danger score to different types of road users. Method according to one of the preceding claims, characterized by - gradually updating the data included in the digital mapping of traffic route networks, in particular the danger scores, when a predetermined amount of further sensor data has been collected and/or a predetermined amount of further user input data has been collected; and or - through adapted data collection of geo-referenced sensor data, related to the movement behavior of a road user, in particular with increased sensitivity or intensity, when detecting the road user's local proximity to a segment that has already been identified as a potential danger spot. Computer system, in particular cloud computer system, comprising a map-like digital traffic route network image, which has segments with a danger score identified as danger spots according to a method according to one of the preceding claims 1 to 13. Method for generating danger zone-related warning messages to road users, with terminal devices communicating with the computer system according to claim 14, in particular smartphones and / or navigation devices, issuing a warning message when local proximity to a segment identified as a danger spot is detected, in particular a warning message comprising a relevance information and / or danger type information and/or a contextual statement. Use of the computer system according to claim 14 - to determine route options that are safe in terms of traffic in a navigation system, in particular in addition to fast and/or ecological route options determined by the navigation system; or - for communication with autonomous or semi-autonomous driving vehicles in order to influence autonomous or semi-autonomous driving behavior depending on danger spots, in particular taking into account relevance and/or danger type and/or context information and/or for the purpose of influencing route selection; or - for determining and/or optimizing the ODD of an autonomously driving vehicle (AV), in particular an autonomously driving vehicle of the L3+ type, in particular in an unlimited ODD of an L5-AV; or - to provide a data set for training a machine learning model to predict potential danger spots, the machine learning model being used in particular to optimize the map-like digital traffic network mapping and/or in a simulation for traffic planning.

Images