Computer-implemented method and system for analyzing driving data of own vehicle

Abstract

A computer-implemented method for analyzing driving data of an ego-vehicle, the driving data including trajectory data of the ego-vehicle along a test route and trajectory data of other traffic participants of an environmental area surrounding the test route. The method entails a predefined set of abstract scenes, a digital map covering the test route and surrounding areas, on which at least the respective topological areas of the test route area map are geometrically located, and an analysis model for the test route area map for determining an equivalence class of the own vehicle behavior on the basis of the areas. According to the invention, a sequence of regions is thus determined for each participant. Then, on the basis of the sequence of regions of the participants, a sequence of equivalence classes of the behavior observed by the host vehicle and a sequence of phases differing in terms of the occupancy of the region by the at least one participant and / or the equivalence classes of the behavior observed by the host vehicle are determined by means of an analysis model. Finally, based on the sequence of phases, the behavior observed by the own vehicle is assigned to the provided abstract scene.

Description

Background Technology

[0001] Autonomous and highly automated driving functions should ensure safe and correct driving behavior under various traffic conditions. This requires extensive testing using simulated or recorded driving data, representing the autonomous vehicle's movement along a test route and the surrounding environment where other traffic participants are present. It is essential to ensure that the driving data covers as many different scenarios as possible, especially critical traffic conditions that rarely occur in reality. To this end, abstract test scenarios can be defined, which declaratively describe the progression of a scenario in a manner that, for example, requires a specific sequence of events but does not predefine how these events should occur. Descriptions of such abstract scenarios typically include qualitative information about traffic infrastructure, such as "occurring at a four-way intersection" or "on a crosswalk," without referencing specific geometric maps. Therefore, many logical scenarios can be assigned to an abstract scenario, all of which satisfy the abstract description of the scenario but differ from one another due to different geometric maps, different traffic participant situations, and / or permissible differences in specific implementations.

[0002] This invention relates to computer-implemented self-driving data analysis for vehicles, which should be used to test autonomous and / or highly automated driving functions. In particular, the extent to which the driving data covers a pre-given set of abstract scenarios should be examined.

[0003] The driving data includes at least the trajectory data of the self-vehicle along the test route and the trajectory data of at least one other traffic participant in the surrounding environment of the test route, wherein the trajectory data includes the location data of the corresponding participant (self-vehicle or other traffic participant) for a trajectory time sequence. The analysis of the driving data is based on a set of abstract scenarios. Furthermore, a digital map is provided that covers the test route and its surrounding environment, and at least the various topological regions of the zongraph of the test route are geometrically located on the digital map. Additionally, an analytical model for the zongraph of the test route is provided. This analytical model enables the determination of equivalence classes for the self-vehicle's behavior based on the regions.

[0004] Such analytical models are typically used for region-based behavioral analysis of autonomous vehicles within the scope of prediction and planning. The core idea of ​​region-based behavioral analysis is to understand the possible intentions of an autonomous vehicle in a traffic scenario as a sequence of topological driving zones. For each driving zone, conflicting traffic flows are identified and represented as so-called location zones. Furthermore, important elements of traffic infrastructure, such as traffic signs and traffic lights, are represented through so-called information zones, where the presence and status of these important elements must be known to the autonomous vehicle. The resulting topological region map is abstracted from the specific geometric information and conditions of the actual traffic infrastructure, and therefore possesses a degree of universal validity.

[0005] The so-called SOCA method was proposed in the paper "SOCA: Domain Analysis for Highly Automated Driving Systems" by M. Butz et al., 2020 IEEE 23rd International Conference on Intelligent Transportation Systems (ITSC), pp. 1-6, doi: 10.1109 / ITSC45102.2020.9294438. The goal of analyzing traffic conditions using the SOCA method is to determine the boundary conditions or requirements for the behavior of automated self-driving vehicles under given traffic conditions. To this end, an abstract description of the traffic conditions to be analyzed is first generated, using a zone graph. The zone graph abstracts the traffic conditions by representing the actual road conditions in different zones using corresponding abstract traffic infrastructure elements (static road geometry). These different zones are related to the self-driving vehicle's driving intentions, but their size and location are initially not specified. These different zones can represent different map areas, possible traffic flows, objects, etc. Based on an abstract description of the traffic conditions, the possible developments or behaviors of the traffic participants involved are identified and morphological analysis is performed to determine the boundary conditions of ego vehicle behavior under the analyzed traffic conditions. It is worth noting that the results or boundary conditions obtained in this way are initially applicable to all traffic conditions with the same area map. Specific details can only be specified by loading the results, including condition-specific parameters for the analyzed traffic conditions.

[0006] In the driving data analysis discussed here, the test route of the ego vehicle and the underlying geometric map are pre-given by the driving data to be analyzed. Based on this information, the various regions of a region map (typically used for region-based behavior analysis) can be defined and geometrically located on a digital map. This region map then represents a specific representation of the topological region map. That is, an instance (Instanz).

[0007] The provided analytical model can be, for example, a so-called "Zwicky box," as used within the scope of the SOCA method for behavior analysis. Zwicky boxes specifically model the external influences (e.g., traffic light status) and system states (e.g., speed) that a vehicle must consider in making behavioral decisions. Using Zwicky boxes, equivalence classes describing what behavior the vehicle should exhibit under what conditions are identified. The SOCA method guarantees the integrity of the decision space. Furthermore, they are consistently (konsistently) assigned to equivalence classes. In this regard, "complete" means that every possible combination of external influences and system states is assigned to at least one equivalence class, while "congruent" means that every possible combination of external influences and system states is assigned to at most one equivalence class. Summary of the Invention

[0008] According to the present invention, it has been recognized that region-based behavior analysis, such as the SOCA method, can also be used to analyze test data for automated driving functions, given the coverage of pre-given test scenarios. By means of region-based behavior analysis, the abstract scenario on which the recorded driving data is based can be determined. Therefore, it is possible to determine which abstract scenarios are covered by the test data set.

[0009] According to the invention, firstly, a region sequence is determined for each participant, that is, the sequence of regions in a digital map traversed by the corresponding participant (the self-vehicle and at least one other participant). For this purpose, the digital map and trajectory data of each participant contained in the recorded driving data are used. Then, a sequence of equivalence classes for the observed behavior of the self-vehicle is determined. For this purpose, the region sequence of the self-vehicle is analyzed using an analytical model, taking into account the region sequences of other participants. Therefore, a sequence of phases (Phasen) for the self-vehicle's journey is then determined. The phases in this sequence differ in terms of at least one participant's occupancy of a region and / or the equivalence class of the observed behavior of the self-vehicle. Based on the phase sequence determined in this way, the observed behavior of the self-vehicle (i.e., the recorded journey of the self-vehicle) is assigned to at least one of the provided abstract scenarios.

[0010] In a preferred embodiment of the invention, the trajectory data is evaluated and processed point by point, that is, the location data of the corresponding participant is analyzed and processed separately for each trajectory time step. By determining which area of ​​the digital map the participant was in based on the corresponding location data for each trajectory time step, the sequence of areas for that participant can be determined very easily in this case.

[0011] Advantageously, the analysis of the regional sequences is also performed point by point. Here, for each trajectory time point, an equivalence class is determined for the observed behavior of the self-vehicle.

[0012] Therefore, it is easy to combine the sequence of regions of each participant and the sequence of equivalence classes of observed behavior toward the self-vehicle point by point to determine the sequence of phases that differ in at least one participant’s occupancy of a region and / or the equivalence class of observed behavior toward the self-vehicle.

[0013] Typically, in addition to participant trajectory data, the driving data to be analyzed usually includes other information about the observed traffic scene, such as traffic light status, road conditions, and weather conditions. Advantageously, the analytical model considers this additional driving data when determining the sequence of equivalence classes for the observed behavior of its own vehicle.

[0014] As previously stated, the analysis according to the invention can determine whether the recorded driving data covers a predetermined abstract scenario. In particular, it is advantageous if the invention is intended to be used within the scope of ensuring autonomous or highly automated driving functions during the design phase, as the provided set of abstract scenarios meets a pre-defined integrity criterion.

[0015] If the abstract scenario is described at the topological level based on a topological region of a region graph, then existing automata-based coverage measures, such as state coverage, transition coverage, or path coverage, can be used as integrity criteria. Thus, for example, similar to state coverage in the case of automata, stage coverage can be required such that each possible stage appears at least once in any abstract scenario of the given set. Similarly, similar to path coverage in the case of automata, it can be required that the abstract scenarios of the given set cover every possible sequence of stages.

[0016] In one embodiment of the invention, at least a portion of the provided abstract scenarios is determined as a defined sequence of stages using an analysis model. In this case, the sequence of stages determined in the driving data analysis can be easily compared with a predefined sequence of stages for the provided abstract scenarios to check whether the driving data covers one of the pre-given abstract scenarios, and if so, to check which one is specifically covered.

[0017] Another possibility is to provide an abstract set of scenarios in the form of a generative model of the corresponding scenario space. This offers the advantage of not needing to explicitly enumerate possible scenarios. In this case, to assign the observed behavior of the ego vehicle to at least one scenario in the provided scenario space, it is checked whether the phase sequence determined in the analysis of the driving data is "included" in the generative model. For this purpose, the generative model needs to be converted into a monitor form so that so-called member testing can then be performed.

[0018] Using the method according to the invention, the suitability of a set of driving data for testing automated driving functions can be reliably evaluated, provided that the test should cover a pre-given set of abstract test scenarios. To this end, the ratio (ins) of the set of "seen" abstract scenarios determined within the scope of driving data analysis according to the invention to the total set of provided abstract scenarios is determined. (gesetzt). The size of the subset of "seen" scenarios is used here as the basis for evaluation of the driving data set.

[0019] A particular advantage is the ability to record and analyze driving data during the operation of the vehicle itself. In this case, when recording driving data, for example, during test drives and continuous operation, coverage of a pre-defined set of test scenarios can be determined. Furthermore, the invention can also be used to analyze fleet data.

[0020] In addition to the aforementioned computer-implemented method for analyzing self-driving vehicle data, the present invention also relates to a computer-implemented system for analyzing self-driving vehicle data. According to the present invention, such a system includes at least one storage medium for:

[0021] ○ The vehicle's driving data, wherein the driving data includes at least the trajectory data of the vehicle along the test route and at least the trajectory data of at least one other traffic participant in the surrounding environment of the test route.

[0022] ○ Abstract scene collection,

[0023] ○ A digital map, which covers the test route and the surrounding environment, and at least the various topological regions of the test route's area map are geometrically located on the digital map, and

[0024] ○ An analytical model for the area map of this test route, used to determine equivalence classes for self-vehicle behavior based on the area. And...

[0025] Furthermore, according to the present invention, such a system includes at least one evaluation module for:

[0026] ○ Using digital maps and based on corresponding trajectory data, a sequence of areas is determined for each participant (the vehicle itself and other traffic participants).

[0027] ○ By using analytical models and based on participants' regional sequences, equivalence class sequences of observed behaviors for self-vehicles are determined.

[0028] ○ Identify the sequence of stages that differ in equivalence classes of at least one participant's occupancy of the area and / or observed behavior toward their own vehicle, and

[0029] ○ Assign the observed behavior of the self-vehicle to at least one of the provided abstract scenarios based on the phase sequence. Attached Figure Description

[0030] The embodiments and advantageous extensions of the present invention will now be described in more detail with reference to the accompanying drawings.

[0031] Figure 1 A block diagram of a first embodiment of a computer-implemented method for analyzing driving data of a self-owned vehicle according to the present invention is shown.

[0032] Figures 2a to 2c A Venn diagram is shown to illustrate the results of the analytical method according to the invention in the form of a set of abstract scenarios described by a generative model. Detailed Implementation

[0033] With the help of this invention, it is possible to automatically determine what kind of abstract scenario the specific scenario described by the driving data is based on. Figure 1 The sequence of method steps according to the present invention is explained, and in particular, it is explained what information (input) is required for each method step and what information (output) is generated in each method step.

[0034] The applicability of the method according to the invention is contingent upon the following condition: the driving data 10 to be analyzed includes: trajectory data of the self-vehicle along the test route, and trajectory data of at least one other traffic participant in the surrounding environment of the test route. The trajectory data typically includes position data of the corresponding participant (self-vehicle or other traffic participant) for a sequence of trajectory time points, but may also include other state data of the corresponding participant at each trajectory time point, such as speed, orientation, etc. In addition to the trajectory data, the driving data 10 to be analyzed typically also includes other data describing the observed traffic scene, such as information about traffic infrastructure, road conditions, weather conditions, etc.

[0035] The method according to the invention requires a digital map 20 that covers the test route of the vehicle along with the surrounding environment, and at least the various topological regions of the area map of the test route are geometrically located on the digital map. For example, such a digital map can be created using the method described in German patent application 102023209189.5.

[0036] Furthermore, an analysis model 30 must be provided for the area map of the test route to determine equivalence classes for the self-vehicle's behavior based on the area. In the embodiments of the invention described herein, the provided analysis model is a so-called "Zwicky box," similar to those used in SOCA methods for behavior analysis. However, in principle, it is also possible to start from another form of model that describes the scene space as a state machine.

[0037] Finally, the method according to the invention requires a set of abstract scenarios that should be identified in the driving data to be analyzed.

[0038] In the method variant described herein, such abstract scenarios are predetermined in method step 101. To this end, the traffic scenario predetermined by the driving data 10 is analyzed using analysis model 30. Here, a predetermined integrity metric 40, such as state coverage, path coverage, etc., is considered. To determine the abstract scenario using analysis model 30, the method described in German patent application 102020207909.9 can be used, for example. This method and the coverage metric determine possible abstract scenarios for the existing geometry. The abstract scenarios determined in method step 101 are stored in a database.

[0039] According to the method of the present invention, a sequence of areas is determined for each participant using a digital map 20 and based on trajectory data 10. This is performed in method step 102. In method step 102, for each trajectory time point contained in the driving data 10, it is determined not only for the participant's own vehicle but also for other participants: which driving area the participant's own vehicle is located in, and which other participants have been located in which driving areas / location areas. Since other participants may of course remain in the participant's own vehicle's driving area, driving areas should always be considered as location areas. For this purpose, information about the geometric location of areas in the digital map is utilized. Thus, a sequence of areas is obtained for each participant in the form of occupied areas at the corresponding trajectory time point.

[0040] According to the invention, the sequence of regions determined for each participant in method step 102 is analyzed in method step 103 by means of an analytical model to determine a sequence of equivalence classes for the observed behavior of the self-vehicle. Here, other information contained in the driving data is also evaluated. Therefore, the Zwicky box used here as the analytical model determines the applicable equivalence classes for the behavior analysis. Depending on the form of the behavior analysis, either a required behavior or multiple permitted behaviors can be derived for the self-vehicle. The evaluation is performed for each time step, thereby obtaining a sequence of equivalence classes for each trajectory time point. This sequence of equivalence classes is now combined point-by-point with the region sequence of the participants determined in method step 102 to obtain a stage for each trajectory time point. To obtain a sequence of stages that differ in terms of at least one participant's occupancy of a region and / or the equivalence class of the observed behavior for the self-vehicle, all consecutive identical stages are now removed, so that between every two stages, at least one region occupancy of at least one traffic participant changes or the equivalence class of the self-vehicle changes.

[0041] In method step 104, based on the phase sequence determined in method step 103, the observed behavior of the ego vehicle is assigned to at least one abstract scenario provided in method step 101. To do this, the scenario identified by the phase sequence is searched in the database created in method step 101. To speed up the search, techniques such as hashing can be used. In method step 104, situations may arise where it is impossible to clearly distinguish or assign individual scenarios. In such cases, both scenarios can be marked as identified, and the data trajectories can be labeled as instances of the two abstract scenarios respectively.

[0042] In the embodiments described herein, all scenarios seen in the driving data 10 are finally aggregated in method step 105 to determine which abstract scenarios of the set determined in method step 101 are covered (labeled 50 here) and which abstract scenarios of that set are not covered (labeled 60 here). Additionally, the coverage 70 for the analyzed driving data 10 can be determined based on the ratio of these two sets 50 and 60. In the embodiments described herein, this only requires labeling and counting the scenarios contained in the database.

[0043] Unlike step 101 of the method described above, the abstract set of scenarios can also be provided in the form of a generative model of the corresponding scenario space, such as a regular language or a nondeterministic finite automaton. These two representations are equivalently convertible to each other. The advantage of this is that it is not necessary to explicitly enumerate all possible scenarios. In this case, it is important that the regular language is either finite or that a maximum length k must be additionally defined for these scenarios, where k represents the maximum word length of the regular language. Otherwise, coverage cannot be achieved or the algorithm will not terminate. Figure 2a Abstract scenario A is shown. Scenario The Venn diagram of the set described in this way for 200.

[0044] In this case, in method step 104, it must be checked whether the scenes seen in the analyzed driving data are "contained" in the generative model. To do this, the generative model is converted into a monitor form, and the seen scenes are then input into the monitor stage by stage as traces. The monitor then performs nondeterministic matching of the traces relative to a regular language / automaton, also known as membership testing. Here, one or more possible scenes from the scene space are provided. The latter is particularly useful when two possible equivalence classes cannot be precisely distinguished due to a lack of information.

[0045] To summarize the seen scenarios in step 105, an automaton needs to be learned from all the seen scenarios. This is done by interpreting all seen scenarios in the analyzed driving data as trajectories of an unknown automaton. Now, using these trajectories, an automaton A describing these trajectories is learned. Trace For this purpose, known methods such as the L* algorithm can be used, for example, see: Angloin, Dana, “Learning regular sets from queries and counterexamples”, Information and computation 75.2 (1987): 87-106. Figure 2b It shows the result of A Trace The description of the collection of scenes that have been seen 210 and A Trace The complement of 211 is the Venn diagram.

[0046] To determine the coverage of 70, first determine the automaton A to be learned. Trace The complement of 210 is 211, and then regular language A is used. Scenario Calculate the intersection of A and B. uncov 260, that is Figure 2c The calculation formula is presented graphically as a Venn diagram, because automata A Trace 210 and A Scenario 200 can also be understood as a set of scenarios they describe. If The driving data being analyzed then covers A Scenario All abstract scenarios defined by 200. If not, by A. uncov All implementation scenarios described in section 260 are not yet covered. uncov The existence of 260 and A uncov The fact that 260 is also a regular language has been proven, for example, see the lecture notes: https: / / homepage.cs.uri.edu / faculty / hamel / courses / 2014 / spring2014 / csc445 / lecture-notes / ln445-03.pdf.

[0047] Here, it should also be noted that, with the help of A uncov Optionally, additional test data can also be generated, covering scenarios that have not yet been covered.

[0048] In another embodiment of the invention, the formal language L Scenario Generative models of this form are also used as a basis for describing possible scenarios, i.e., for providing a set of abstract scenarios. Scenaro It can also be a regular language, but other formal languages ​​defined by an alphabet (Alphabet) can also be used, such as context-free grammars (kontextfreieGrammatiken).

[0049] In this implementation, the possible stages that occur in the scene model are considered as formal language L ScenarioThe letters in the alphabet. This is used in method step 104 when identifying or assigning seen scenes, by iteratively traversing the seen trajectories t in the driving data and determining, for example, by calculating the Brzozowski derivatives (https: / / en.wikipedia.org / wiki / Brzozowski_derivative): based on the prefix p processed up to time point i. 0..i Which scenarios are still possible? The set of still possible scenarios is denoted by S. If the set S is empty, then the scenario is not included in the model and represents a new scenario. Once only one possible scenario remains, i.e., |S| = 1, a comparison can be made: whether the suffix in s ∈ S corresponds to the suffix of t. If so, the scenario has been successfully found.

[0050] The scene space generation model is also used here to aggregate all seen abstract scenes by progressively generating all possible prefixes of a scene of length i, i.e., the beginning of an abstract scene with i stages. For each p in the list of seen scenes... i Check: Does the recorded trajectory contain a character prefixed with 'p'? i The opening scene i If not, then s i All possible continuations (Fortsetzungen) are scenarios that have not yet been covered. Their number can optionally be determined by using the prefix p. i In the case of formal language L Scenario Perform exhaustive enumeration And thus, (geschehen). If only the uncovered scenes should be described, then the Brzozowski derivative can be used again, for example.

[0051] In another embodiment of the method according to the invention, all possible abstract scenarios are first listed and stored in a database. Then, method steps 102 and 103 are performed, as in combination. Figure 1 As stated above.

[0052] However, in order to identify the observed behavior of the self-vehicle or to assign this behavior to at least one of the provided abstract scenarios, the conditions of each stage of the abstract scenario are transformed into atomic propositions or predicate logic formulas. Formeln), these propositional or predicate logic formulas can be evaluated as "true" or "false" on the trajectory. Here, atomic propositions can refer to both zone maps and Zwicky boxes in behavioral analysis. Examples of atomic propositions are "Fahrzeug-in-zone-X (Vehicle in zone X)" and "Ampel_rot (Traffic light_red)". The phase sequence is then encoded into a temporal logic formula, for example, using Signal Temporal Logic (STL), see, for example, https: / / link.springer.com / chapter / 10.1007 / 978-3-319-75632-5_5. Examples of such formulas in STL might look like this:

[0053]

[0054] It should be noted here that it is also possible to use other operators.

[0055] To identify which abstract scenario satisfies the current trajectory, established STL monitoring methods can be used, such as Breach Toolbox (Donzé, A., Ferrer, T., and Maler, O., “Efficient robust monitoring for STL”, July 2013, published at the International Conference on Computer-Aided Verification (pp. 264-279). Springer, Berlin, Heidelberg.).

[0056] This variant of the method has proven particularly advantageous because the evaluation of the timing logic formula is also possible in real time, i.e., it can be performed during the recording of driving data.

[0057] In addition to the above description, the following advantages and extension possibilities of the present invention should also be noted:

[0058] The scenarios to be simulated and / or the requirements for future test drives can be automatically derived from information about uncovered abstract scenarios determined according to the present invention, which makes a valuable contribution to the testing of autonomous driving functions.

[0059] When analyzing driving data according to the present invention, abstract scenarios not included in the provided set of possible abstract scenarios or not described by the generative model can also be identified. Such results can be advantageously used to improve the generative model and / or the underlying analysis model used to describe the possible abstract scenarios.

[0060] Optionally, hypotheses about the behavior of other traffic participants can be specified in the abstract scenario. These hypotheses can be executed in parallel with scenario identification to find scenarios that do not conform to the model hypotheses made. A method for verifying such hypotheses for a region-based model is described in German patent application 102020215545.3, and can be similarly used here.

[0061] Behavioral analysis conducted within the scope of the method according to the invention may also involve only a portion of traffic conditions, such as proper interaction with pedestrians or responses to four-way intersections. In this case, different analytical models may also be used to evaluate multiple sub-situations and... Figure 1 The method step 102 shown is divided into several sub-steps. In the first sub-step, it can be determined which sub-situations should be evaluated for the current trajectory and which traffic participants are relevant. In the following sub-steps, the corresponding analytical models are then used to perform actual behavioral analysis.

[0062] Additionally, these components of traffic conditions may also have dependencies on each other, which can influence possible scenarios. These dependencies can then be decomposed, for example, through prioritization. As described in German patent application 102023201983.3.

Claims

1. A computer-implemented method for analyzing driving data (10) of a self-owned vehicle, wherein, The driving data (10) includes at least the trajectory data of the self-vehicle along the test route and at least the trajectory data of at least one other traffic participant in the surrounding environment of the test route, wherein the trajectory data includes the location data of the corresponding participant (self-vehicle or other traffic participant) for a trajectory time sequence, and In order to analyze the driving data (10), at least the following are provided: -A collection of abstract scenarios (method step 101) - A digital map (20), which covers the test route and the surrounding environment area, and at least the various topological regions of the area map of the test route are geometrically located on the digital map, and - An analytical model (30) for the area map of the test route is used to determine the equivalence class for the behavior of the self-vehicle based on the area; Its features are, - Using the digital map (20) and based on the trajectory data (10), determine the region sequence for each participant (method step 102). - Using the analytical model (30), based on the participant's region sequence, determine the equivalence class sequence for the observed behavior of the self-vehicle (method step 103). - Determine a sequence of stages that differ in terms of the equivalence class of at least one participant's occupancy of the area and / or observed behavior toward the self-vehicle (method step 103), and - Assign the observed behavior of the self-vehicle to at least one of the provided abstract scenarios based on the sequence of the stages (method step 104).

2. The method according to claim 1, characterized in that, The region sequence for each participant is determined by identifying the regions where the participant was located at each trajectory time point (method step 102).

3. The method according to claim 1 or 2, characterized in that, In addition to the trajectory data or region sequence of the participants, a sequence of equivalence classes for the observed behavior of the self-vehicle is also determined based on other driving data (method step 103).

4. The method according to claim 2 or 3, characterized in that, A sequence of equivalence classes for the observed behavior of the self-vehicle is determined by identifying equivalence classes for each trajectory time point (method step 103).

5. The method according to claim 4, characterized in that, The sequence of the phase is determined by combining the regional sequences of each participant with the sequence of equivalence classes of the observed behavior for the self-vehicle, particularly by combining the sequences of individual trajectory time points (method step 103).

6. The method according to any one of claims 1 to 5, characterized in that, The set of abstract scenarios provided satisfies the pre-given integrity criterion (40).

7. The method according to any one of claims 1 to 6, characterized in that, Using the analytical model, at least a portion of the provided abstract scenarios are identified as a defined phase sequence (method step 101).

8. The method according to any one of claims 1 to 6, characterized in that, The set of the abstract scene (200) is provided in the form of a generative model of the corresponding scene space.

9. The method according to claim 8, characterized in that, In order to assign the observed behavior of the self-vehicle to at least one scene in the provided scene space, a member test is performed by means of a monitor form of the generative model.

10. The method according to any one of claims 1 to 9, characterized in that, The driving data set is evaluated based on the analysis by determining a subset of the provided abstract scenario set to which the observed behavior of the self-vehicle can be assigned, and by determining a measure of the extent to which the driving data set covers the provided abstract scenario set based on the subset.

11. The method according to any one of claims 1 to 10, characterized in that, The driving data is recorded and analyzed during the operation of the vehicle.

12. A computer-implemented system for analyzing driving data (10) of a self-owned vehicle, comprising at least: - At least one storage medium, which is used for ● The driving data (10) of the self-vehicle, wherein the driving data (10) includes at least the trajectory data of the self-vehicle along the test route and at least the trajectory data of at least one other traffic participant in the surrounding environment of the test route. ●A collection of abstract scenes ● A digital map (20) covering the test route and the surrounding environment, wherein at least the various topological regions of the area map of the test route are geometrically located on the digital map, and ● An analytical model (30) for the area map of the test route, used to determine equivalence classes for the behavior of the self-vehicle based on the area; and - At least one evaluation module, which is used for ●Using the digital map (20) and based on the corresponding trajectory data (10), a sequence of areas is determined for each participant (self-vehicle and other traffic participants). ● Using the analytical model and based on the region sequence of the participants, a sequence of equivalence classes for the observed behavior of the self-vehicle is determined. ● Identify a sequence of stages that differ in equivalence classes of at least one participant's occupancy of the area and / or observed behavior toward the self-vehicle, and ● Assign the observed behavior of the self-vehicle to at least one of the provided abstract scenarios based on the sequence of the phases.

Images