DE102016216154A1 - Method and evaluation unit for determining the position of traffic signs - Google Patents

Abstract

A method (300) for determining positions of traffic signs (104) in a road network (200) is described. The method (300) comprises determining (303) a path (222) through a subarea (210) of the road network (200). In addition, the method (300) comprises determining (303) measurement data of a plurality of trips of one or more vehicles (101) along the path (222), the measurement data of a trip detecting one or more traffic sign records for corresponding one or more Include traffic signs (104) detected while traveling along the path (222). The method (300) further comprises determining (304), based on the measurement data, a number M (412) of clusters (315, 316) for grouping the traffic sign records and positions (415, 317) of the M clusters (315 , 316) along the path (222), so that an optimization criterion is improved.

Description

The invention relates to a method and a corresponding evaluation unit for determining the position of traffic signs in a road network on the basis of measurement data of a plurality of journeys in the road network.

For the automatic determination of a driving route for a vehicle and / or for the at least partially automated driving of a vehicle, precise information about the position and / or the type of traffic signs is required. The present document is concerned with the technical task of specifically determining the position of traffic signs in a road network in a precise and reliable manner from the measurement data of a plurality of journeys of one or more vehicles in the road network.

The object is solved by the independent claims. Advantageous embodiments are u. a. in the dependent claims.

According to one aspect, a method for determining positions of traffic signs (in particular signs) in a road network is described. The procedure can be executed on a computer or server. In particular, the method is typically a machine-implemented method. The road network may include road sections along which vehicles may travel through the road network. At the edges of a road section and / or on a bridge over a road section traffic signs may be arranged. The road network may be subdivided into subregions, where a subarea may comprise one or more (typically rectangular) cells.

The method includes determining a path through a portion of the road network. In this case, a path may comprise a sequence of one or more road sections which are traversed in a specific direction of travel. The method may be configured to determine the locations of one or more traffic signs along that path. Typically, a path includes a sequence of traffic signs (with multiple traffic signs) that follow one another along a path in a particular order.

The method further comprises determining measurement data of a plurality of trips of one or more vehicles along the path. In the process, the measured data were recorded by means of environment sensors (eg by means of image cameras) of the one or more vehicles. In particular, the one or more vehicles may comprise environment sensors with which sensor data (in particular image data) relating to the surroundings of the one or more vehicles can be detected. The one or more vehicles may determine the sensor data on trips along the path. Based on the sensor data, the measurement data relating to traffic signs along the path can then be determined (eg by analyzing the sensor data, in particular by applying image analysis methods to acquired image data).

The measurement data of a trip includes one or more traffic sign records for corresponding one or more detected traffic signs detected while traveling along the path. In other words, the measured data of a journey along the path typically include a sequence of traffic sign data sets for a corresponding sequence of detected traffic signs. In this case, a traffic sign data set (possibly exactly) for a detected traffic sign indicates position data relating to a position of the detected traffic sign. In particular, a traffic sign data set may indicate the position of the vehicle on which a traffic sign has been detected by the environment sensor system of the vehicle. Furthermore, a traffic sign record may also indicate the type of a detected traffic sign. In addition, a traffic sign record may indicate the relative position of a detected traffic sign relative to the position of the vehicle. In other words, the position data may indicate how a detected traffic sign is positioned relative to the vehicle.

Typerweise differ the sequences of traffic sign data sets of different trips from each other. This is due in particular to measurement tolerances and / or measurement errors of the surroundings sensor system of the one or more vehicles. Furthermore, the vehicles typically have different states (eg different speeds of travel) during different journeys along the path, which can lead to different measurement results. In addition, the weather conditions may differ on different trips and thus affect the measurement results. Measuring errors can lead to the fact that an actual traffic sign is not detected or that there are misdetections of a non-existent traffic sign. with sum can thus be noted that the measurement data for determining the positions of traffic signs along the path are typically faulty.

The method further comprises determining, based on the measurement data, a number M of clusters for grouping the traffic sign records. M is an integer. The number M of clusters can be greater than zero, in particular greater than one. Furthermore, based on the measurement data, positions of the M clusters along the path are determined. The number M and the positions are determined in such a way that an optimization criterion is improved (in particular maximized). In this case, a cluster (in particular a so-called path cluster or supercluster) indicates at least (possibly exactly) a traffic sign, and the position of a cluster indicates the position of the at least one traffic sign.

By considering sequences of traffic sign records for a sequence of detected traffic signs along a path, the reliability and accuracy of the determined locations of traffic signs (ie, clusters) can be increased (particularly as compared to individually viewing the traffic sign data sets of individual detected ones traffic signs). In particular, it can be achieved that the increase in the number of journeys, ie. H. increasing the number of measured traffic sign records does not result in (typically incorrect) increases in the number of detected clusters and cluster locations.

The optimization criterion may depend on a probability (referred to in this document as p ₂ ) for a traffic sign to be detected when driving along the path from the environment sensor system of a vehicle. In other words, the optimization criterion may depend on a probability that a traffic sign is detected by the environment sensor system of any vehicle. Thus, it can be achieved that the number M of the determined clusters with increased accuracy corresponds to the actual number of actual traffic signs along the path.

The path includes (typically multiple) actual traffic signs at actual positions represented by the measurement data. In other words, the actual traffic signs along the path lead to the measurement data. The optimization criterion may be designed such that the optimization criterion indicates the probability that the determined positions of the M clusters correspond to the actual positions of the actual traffic signs along the path. For example, the optimization criterion for this purpose may include a log likelihood criterion. Thus, the positions of the traffic signs along the path can be determined with high accuracy or quality.

The optimization criterion can be a probability (in this document referred to as p ₁₎ depend on that it is in a detected traffic sign when traveling along the path to a false detection. Thus, the quality of the determined positions of the traffic signs can be further increased.

The optimization criterion may depend on a likelihood model that indicates the likelihood of detecting a traffic sign in response to a state of a vehicle as it travels along the path. A traffic sign record may then indicate the state of the vehicle upon detection of the traffic sign record. Thus, the state of a vehicle detecting the measurement data can be taken into account in the determination of the position of a traffic sign in order to further increase the quality of the determined position. The state of the vehicle may in particular depend on a speed and / or a rate of rotation of the vehicle (eg during a turning operation).

Determining the locations of the M clusters may include assigning the traffic sign records of the plurality of trips to the M clusters. The assignment can be made in such a way that a cluster is assigned a maximum of only one traffic sign record of a trip of the plurality of trips. This condition may apply to each of the plurality of trips. Furthermore, the assignment can take place such that a traffic sign data record is assigned to only one cluster. Thus, the fact can be taken into account that a vehicle only detects an actual traffic sign just once when driving along the path. By taking into account such boundary conditions, the quality of the determined positions can be further increased.

Determining the locations of the M clusters may include determining the location of a cluster based on the location data of the one or more traffic sign records associated with the cluster. Thus, the position of a cluster (and a traffic sign displayed thereby) can be determined in a precise manner.

The determination of the positions of the M clusters can be done iteratively. In particular, on the basis of the determined cluster positions, a reassignment of the traffic sign data records can take place, and new cluster positions can be determined based on the reassignment. This iterative process can be repeated until an abort condition is reached in order to determine the (optimized) positions of the M clusters. The termination condition can z. B. depending on a convergence of a value of the optimization criterion. This allows the positions of the clusters to be determined in a reliable and precise manner.

Determining the number M of clusters may include determining an (optimized) value of the optimization criterion for different numbers of clusters. The (optimal) number M can then be determined on the basis of the values of the optimization criterion for the different numbers of clusters. In particular, the number M for which the relatively best value of the optimization criterion was determined can be selected. Thus, the number of traffic signs along the path can be determined in a precise manner.

The optimization criterion may depend on a probabilistic model for measurement errors in the position data with respect to the position of the detected traffic sign. In particular, when assigning traffic sign data sets to clusters, such a model can be taken into account for measurement errors in the position data. Thus, the clusters and the positions of the clusters can be determined with increased accuracy.

Alternatively or additionally, the optimization criterion may depend on a distance between the determined position of a cluster and the positions of one or more detected traffic signs whose one or more traffic sign data sets have been assigned to the cluster. In other words, the optimization criterion can take into account the distance between the determined position of a cluster and the positions of the detected traffic signs associated with the cluster. In particular, relatively large distances can be punished. Thus, the clusters and the positions of the clusters can be determined with high accuracy.

mitIn a preferred case, the optimization criterion for a path p (or corresponds to the optimization criterion) comprises:

With

The optimization criterion LLH p / path may possibly be a subset of the above terms LLH i / sign messages . LLH i / path cluster and LLH p / Prior include (eg LLH i / sign messages and LLH i / path cluster ).

Where n _{passes are} the number of trips along the path p. ni / sign messages corresponds to the number of traffic signs records from the measurement data for the i ^th driving. LLH ij / assignment indicates the likelihood, in particular the log likelihood, that the traffic sign data set j of the i ^th journey corresponds to one Cluster is assigned or assigned. It can LLH ij / assignment depend on a distance between a position of the detected traffic sign of the traffic sign data set j of the i ^th drive and the determined position of a cluster to which the traffic sign data set j of the i ^th drive is assigned. Alternatively or additionally LLH ij / assignment correspond to an empirically determined value if the traffic sign data set j of the i ^th drive is or will not be assigned to a cluster. δ _ij assumes the value 1 or 0, depending on whether the traffic sign data set j of the i ^th drive is assigned to a cluster or not. p ₁ can indicate a probability that a traffic sign record is a hoax. np / path cluster corresponds to the number of clusters on the path p. After performing the procedure corresponds np / path cluster typically the optimal number M of clusters. ε _ik assumes the value 1 or 0, depending on whether or not the cluster k has been assigned a traffic sign data set for the ^ith trip. p ₂ may indicate a probability that an actual traffic sign on the path p is detected when traveling along the path p. n _Prior can correspond to a (fixed) number of virtual trips in which no traffic signs are detected.

When using the o. G. Optimization criteria, the number M of clusters and the positions of the clusters can be determined in a particularly precise manner.

The method may include determining the positions of clusters, hereinafter referred to as path clusters, for a first path and for an at least partially different second path through the subarea of the road network. In this case, the first path may comprise a first road section and the second path may comprise a second road section of the road network, wherein the first road section and the second road section may be designed so that a traffic sign on the first road section can be detected when driving along the second road section. This can be z. B. occur when the first road section and the second road section parallel to each other.

It can then be determined based on the path clusters for the first and second path, a position of a supercluster for the first road section, so that a supercluster optimization criterion is improved. The supercluster indicates at least one traffic sign and the position of the supercluster indicates the position of the at least one traffic sign. The supercluster optimization criterion may depend on a likelihood that a traffic sign of a road segment may be detected while driving along another road segment.

By identifying super clusters for individual road sections based on the previously determined path clusters for different paths, overlaps between the different paths can be detected and evaluated in a reliable and precise manner. Thus, the quality of the determined positions of traffic signs can be further increased.

The supercluster optimization criterion may depend on a plausibility of the ascertained position of the supercluster for the first road segment with respect to typical locations of traffic signs in a road segment. Typical positions of traffic signs in a road section can be determined empirically. Thus, the accuracy of the positions of superclusters (or associated traffic signs) can be increased.

Alternatively or additionally, the supercluster optimization criterion may depend on a distance between the determined position of the supercluster and the positions of the one or more path clusters associated with the supercluster. This allows the positions of superclusters (and therefore traffic signs) to be determined in a precise manner.

mit

In a preferred example, the supercluster optimization criterion comprises (or corresponds to the supercluster optimization criterion:

With

The optimization criterion LLH _Global may contain a subset of the above terms LLH i / Path Cluster, LLH i / Supercluster and LLH _{constellation} include (e.g. LLH i / path cluster and LLH i / Supercluster ).

Here, n _paths of the number of paths that are considered for the determination of superclusters corresponds. ni / path cluster corresponds to the number of path clusters of the i ^th path. LLH ij / assignment, supercluster indicates the probability, in particular the log likelihood, that the path cluster j of the ^ith path is assigned to a supercluster. It can LLH ij / assignment, supercluster depend on a distance between a position of the path cluster j of the ^ith path and the ascertained position of a supercluster associated with the path cluster j of the ^ith path. Alternatively or additionally LLH ij / assignment, supercluster correspond to an empirically determined value if the path cluster j of the i ^th path is not assigned to a supercluster. δ ^S _ij takes the value 1 or 0, depending on whether the path cluster j of the ^ith path is assigned to a supercluster or not. p ij / 1 may indicate a probability that the one or more traffic sign records associated with the path cluster j of the ^ith path are false reports. n _Supercluster corresponds to the number of super clusters to be determined (and can be varied and optimized within the scope of the procedure). ε ^S _ik assumes the value 1 or 0, depending on whether the supercluster k has been assigned a path cluster of the i ^th path or not. p ik / 2 may indicate a probability that a traffic sign of the supercluster k is visible from the i ^th path and is detected. LLH k / cluster position can indicate a probability, in particular a log likelihood, of the ascertained position of a supercluster for the road section to which the supercluster has been assigned. LLH _{configuration} may indicate a likelihood of a particular log-likelihood, the association between Supercluster and road section.

With the o. G. Optimization criterion, overlaps between the path clusters of different paths can be reliably and precisely detected and evaluated in order to determine with high accuracy the positions of traffic signs.

The method may include determining the positions of M _l path clusters for L different paths, where L is an integer, with L> 0, in particular L> 1, 2, 5, 10, 100 (typically n _paths = L). M _l corresponds to the number of path clusters on the l ^th path, with l = 1,..., L (typically ni / path cluster = M _l for i = 1). The method may include determining, based on the path clusters of the L different paths, an (optimal) number N of superclusters so that the supercluster optimization criterion is improved. Then the (optimal) positions for the N super clusters can be determined. Thus, the accuracy of the determined positions of traffic signs can be further increased.

The partial area for determining a path comprises a first cell, which is at least partially enclosed by one or more second cells. For example, the first cell may be completely enclosed by 8 second cells. The method may include discarding a cluster of the M clusters whose location is in a second cell and / or not in the first cell. The method can then be carried out for a further subarea in which one of the second cells is at least partially enclosed by one or more further cells. In order to determine the positions of clusters (ie traffic signs) overlapping subareas can thus be considered. Thus, the quality of the determined positions can be further increased. Furthermore, a decomposition of a road network into subareas allows a linear scaling of the method to arbitrarily large road networks.

In the case of discarding, the possibility can be provided that a cluster is not discarded even if the determined position lies outside the first cell. In particular, such a cluster can be taken into account if the method is carried out for a further subarea. In particular, such a cluster can be taken into account in the determination of super clusters. Thus, the reliability of the detection of traffic signs can be further increased.

The identified clusters (ie traffic signs) and the identified positions can be used to improve a digital map of the road network. In particular, the determined information can be used in the context of the determination of routes through the road network. Furthermore, the information for the (possibly fully autonomous) driving a vehicle through the road network can be used.

According to a further aspect, an evaluation unit for determining positions of traffic signs in a road network is described. The evaluation unit (eg a server or computer) is set up to determine a path through a subarea of the road network. In addition, the evaluation unit is set up to determine measurement data of a plurality of journeys of one or more vehicles along the path, the measurement data having been acquired by means of environment sensors of the one or more vehicles. The measurement data of a trip comprises one or more traffic sign records for corresponding one or more detected traffic signs detected while traveling along the path, wherein a traffic sign record for a detected traffic sign comprises position data relating to a position of the detected traffic sign. The evaluation unit is further set up on the basis of the measurement data to determine a number M of clusters for grouping the traffic sign data sets and positions of the M clusters along the path, so that an optimization criterion is improved. A cluster indicates at least one (possibly exactly one) traffic sign and the position of a cluster indicates the position at least (possibly exactly) of a traffic sign. The optimization criterion may depend on a probability that a traffic sign is detected by the environment sensor of the one or more vehicles when driving along the path.

In another aspect, a software (SW) program is described. The SW program can be set up to run on a processor and thereby perform the method described in this document.

In another aspect, a storage medium is described. The storage medium may include a SW program that is set up to run on a processor and thereby perform the method described in this document.

It should be understood that the methods, devices and systems described herein may be used alone as well as in combination with other methods, devices and systems described in this document. Furthermore, any aspects of the methods, devices, and systems described herein may be combined in a variety of ways. In particular, the features of the claims can be combined in a variety of ways.

Furthermore, the invention will be described in more detail with reference to exemplary embodiments. Show

1 an exemplary measuring vehicle for acquiring measured data;

2 an exemplary road network;

3 a flowchart of an exemplary method for determining the position of traffic signs;

4 a flowchart of an exemplary method for determining path clusters along a path in a road network; and

5 a flowchart of an exemplary method for aggregation of path clusters.

As set forth above, the present document is concerned with the accurate and reliable determination of the positions of traffic signs (also referred to in this document by way of example as signs) in a road network based on measurement data of a plurality of trips of measuring vehicles in the road network. 1 shows an exemplary measuring vehicle 101 with a camera sensor 102 that have an area 103 in front of the vehicle 101 detected. A traffic sign 104 is from the camera sensor 102 captured and it can be the type of traffic sign 104 (Speed limit 60 km / h in the example shown) are determined. It can measure data related to a detected traffic sign 104 to be provided. The measurement data can be the type of traffic sign 104 Show. Furthermore, the measurement data can indicate the position at which the vehicle is 101 in the detection of the traffic sign 104 has found. From the position of the vehicle 101 and the coverage area 103 can then to the position of the traffic sign 104 getting closed. In addition, the measured data can indicate at which position 105 within the coverage 103 the traffic sign 104 located (eg on the right edge 105 of the coverage area 103 ).

• • Information in Bezug auf einen Typ des Verkehrszeichens 104; und/oder
• • Information in Bezug auf den Zustand (z. B. Position, Geschwindigkeit, Orientierung, Drehrate) des Messfahrzeugs 101 bei der Detektion des Verkehrszeichens 104; und/oder
• • Positionsdaten in Bezug auf einen Position des Verkehrszeichens 104.

In a variety of trips of measuring vehicles 101 In a road network, therefore, measurement data with traffic sign data sets for different detected traffic signs 104 be recorded. In this case, the traffic sign data set for a traffic sign 104 Show:

• • Information related to a type of traffic sign 104 ; and or
• • Information related to the condition (eg position, speed, orientation, yaw rate) of the measuring vehicle 101 in the detection of the traffic sign 104 ; and or
• • Position data related to a position of the traffic sign 104 ,

• • Die Anzahl der ermittelten Schildpositionen hängt typischerweise von der Anzahl der Schildmeldungen (d. h. von der Anzahl von detektierten Verkehrszeichen 104 bzw. von der Anzahl von Verkehrszeichen-Datensätzen) ab. Mit einer steigenden Anzahl von Detektionen von Verkehrszeichen 104 (d. h. mit einer steigenden Anzahl von Verkehrszeichen-Datensätzen) an der gleichen Stelle in einem Straßennetz steigt tendenziell die Anzahl der von einem Clusteralgorithmus angelernten Schilder 104, um die durch die Abweichung zwischen Schildmeldungen (d. h. Verkehrszeichen-Datensätzen) und Schildkandidaten (d. h. ermittelte Cluster-Positionen) entstehenden Optimierungskosten zu minimieren.
• • Eng beieinander stehende Schilder 104 können typischerweise nicht unterschieden werden, da die zugehörigen Meldungen (d. h. Detektionen bzw. Verkehrszeichen-Datensätze) durch einen Clusteralgorithmus typischerweise dem gleichen Schildkandidaten zugewiesen werden und somit bei nah beieinander liegenden Schildern 104 lediglich eine einzige Schildposition (d. h. Cluster-Position) ermittelt wird.
• • Im Rahmen eines Clusteralgorithmus auf Basis der individuellen Schildmeldungen bzw. Verkehrszeichen-Datensätze können typischerweise die Geschwindigkeit und die Drehrate eines Fahrzeugs 101 nicht berücksichtigt werden. Eine hohe Geschwindigkeit bzw. Drehrate (z. B. bei einem Abbiegevorgang) kann jedoch zu Ungenauigkeiten bei der Detektion eines Verkehrszeichens 104 führen. Als Folge daraus können sich Ungenauigkeiten bei den erlernten Cluster-Positionen ergeben. Zudem kann es bei hohen Geschwindigkeiten vorkommen, dass die Schildmeldungen (d. h. Verkehrszeichen-Datensätze) zu einem Schild 104 bei unterschiedlichen Fahrten aufgrund von Messungenauigkeiten relativ stark streuen, so dass anstelle eines tatsächlichen Schildes 104 mehrere Cluster-Positionen ermittelt werden.
• • Bei dem Erlernen von Schildpositionen können Heuristiken zu typischen Schildpositionen berücksichtigt werden, um die Genauigkeit eines Clusteralgorithmus zu erhöhen. Dabei ist der Einfluss von Heuristiken zu typischen Schildpositionen auf die erlernten Schildposition typischerweise abhängig von der Anzahl der Schildmeldungen in den Messdaten. Bei Auftreten von systematischen Fehlern in den Schildmeldungen kann dies zum Erlernen von nicht plausiblen Schildpositionen führen.
• • Das Clustern bzw. Gruppieren von individuellen Verkehrszeichen-Datensätzen weist typischerweise einen Rechenaufwand auf, der mit der Größe des berücksichtigten Bereichs eines Straßennetzes überlinear ansteigt, und ist somit typischerweise nicht auf beliebig große Bereiche anwendbar.

The plurality of traffic sign records of a plurality of journeys in a road network may, for. B. can be evaluated individually by a cluster algorithm. However, there are several disadvantages:

• • The number of shield positions determined typically depends on the number of shield messages (ie the number of detected traffic signs 104 or the number of traffic sign records). With an increasing number of detections of traffic signs 104 (ie, with an increasing number of traffic sign records) at the same location in a road network, the number of signs learned by a cluster algorithm tends to increase 104 to minimize the cost of optimization due to the discrepancy between sign messages (ie traffic sign records) and sign candidates (ie, identified cluster locations).
• • Closely spaced signs 104 typically can not be distinguished, since the associated messages (ie, detections or traffic sign records) are typically assigned to the same tag candidate by a cluster algorithm, and thus with closely spaced tags 104 only a single shield position (ie cluster position) is determined.
• • As part of a cluster algorithm based on the individual sign messages or traffic sign data sets typically the speed and the rate of turn of a vehicle 101 not considered. However, a high speed or yaw rate (eg in a turn-off operation) can lead to inaccuracies in the detection of a traffic sign 104 to lead. As a result, inaccuracies in the learned cluster positions may result. Additionally, at high speeds, the sign messages (ie traffic sign records) may become a sign 104 Difficult to scatter on different trips due to measurement inaccuracies, so instead of an actual sign 104 several cluster positions are determined.
• • When learning shield positions, heuristics of typical shield positions can be considered to increase the accuracy of a cluster algorithm. The influence of heuristics on typical shield positions on the learned shield position is typically dependent on the number of shield messages in the measurement data. If systematic errors in the sign messages occur, this can lead to the learning of implausible shield positions.
• The clustering of individual traffic sign records typically has computational overhead that increases linearly with the size of the considered area of a road network, and thus typically is not applicable to arbitrarily large areas.

In the following, a method will be described, in which the traffic sign data sets from a plurality of journeys in a road network are used in a precise and reliable manner to obtain sign positions of traffic signs 104 can be learned in an arbitrarily large area of the road network. By the described method, the above-mentioned disadvantages can be eliminated.

3 shows a flowchart of an exemplary method 300 to determine the positions of signs 104 in an area of a road network 200 (please refer 2 ). The selected area of the road network 200 is also referred to as a job-bounding box in this document 311 designated.

In step S1 (reference character 301 ) are for the given job-bounding box 311 (which, for example, has a size of 2 × 2 km), the relevant driving data (eg position, speed, orientation, turning rate and sign messages) of a plurality of vehicles 101 and passing through the job-bounding box 311 retrieved from a data store, filtered and processed. A timestamp can be used for every signage of a camera system 102 of a vehicle 101 the associated position, orientation, speed and yaw rate of the vehicle 101 be determined by interpolation of the corresponding driving data. In addition, Trace Matching can be the result of a vehicle 101 traffic road sections 221 be comprehended and stored. By processing the information contained in a digital map can be ensured that the considered road sections 221 do not exceed a certain maximum length (eg of 50 meters). Finally, the shield messages and road sections prepared in this way can be used 221 be stored so that they according to their affiliation to individual cells 211 . 212 (with a size of, for example, 100 × 100 m) inside and outside the job-bounding box 311 are sorted. The following are the shield positions of the signs 104 for each of the cells 211 . 212 within the job bounding box 311 calculated separately, in the respective cell 211 . 212 stored and after calculation of the last cell 211 . 212 as a result of the calculation of the job bounding box 311 be issued.

It can thus in a step S1, 301 for a road section 221 a road network 200 and for a passage of the road section 221 a sequence of traffic sign records for a sequence of detected traffic signs 104 be determined during the passage of the road section 221 were detected. This can be for different road sections 221 and for different transits. The road sections 221 can do different cells 211 . 212 of the road network 200 be assigned. This results in a data grid 312 with sequences of traffic sign records for the road sections 221 single cells 211 . 212 ,

For determining the shield positions of a cell 211 become the sign messages and street sections contained therein 221 used. To problems (especially discontinuities) at the edges of a cell 211 In addition, sign messages, road sections can also be avoided 221 and already determined shield positions of the adjacent cells 212 be considered as input data. The entire geographical area of the cells used 211 . 212 for determining the shield positions of a first cell 211 is used in this document as an outer bounding box 210 , the area of the first cell 211 itself referred to as an inner bounding box. The input data may, for convenience, be converted to local coordinates (reference point, for example, the lower left corner of the outer bounding box 210 ) are converted. Thus, in a step S2, 302 local input data 313 for a cell 211 and the directly adjacent cells 212 be determined. The local input data 313 include measurement data with traffic sign data sets for traffic signs 104 that when passing through the outer bounding box 210 were detected.

Based on the input data 313 can in a step S3, 303 possible paths 222 through the outer bounding box 210 and the associated passages (and sequences of traffic sign records) are determined. The basis for this is the result of the trace map matching from step S1, 301 , A path 222 is a sequence of road sections which can be passed through in a defined direction 221 , where the first street section 221 a path 222 none to the outer bounding box 210 belonging, actually passed predecessor and the last section of road 221 a path 222 no corresponding successor in the outer bounding box 210 having. The possibility of turning maneuvers is typically excluded.

In a step S3 303 thus become one or more paths 222 through a cell 211 or by a group of cells 211 . 212 or by a corresponding subarea 210 of the road network 200 determine. This can be different paths 222 at least partially common road sections 221 include. For a path 222 be the sequences of traffic sign records for a variety of passages of the paths 222 determined and as path data or measurement data 314 for the path 222 provided.

mit

wobei n_Überfahrten der Zahl von relevanten Überfahrten (des Pfades p) und n i / Schildmeldungen der Zahl der zu Überfahrt i vorhandenen Schildmeldungen entspricht, und wobei δ_ij = 1 falls die Schildmeldung j der Überfahrt i einem Pfadcluster 315 zugeordnet wurde (sonst δ_ij = 0). LLH ij / Zuordnung entspricht der Log-Likelihood der Schildmeldung j von Überfahrt i gegeben der Position des ihr zugeordneten Pfadclusters bzw. einem von der Fahrsituation abhängigen, empirisch ermittelten Wert, falls kein Pfadcluster zugewiesen wurde. Mit der „Position” eines Schildes 104, eines Pfadclusters 315 oder eines Superclusters 316 kann dessen eigentliche Position und ggf. dessen Orientierung gemeint sein. Das in diesem Dokument beschriebene Verfahren 300 ist entsprechend anwendbar, wenn das jeweilige Pfadcluster 315, Supercluster 316 oder Schild 104 ohne Orientierung modelliert wird.In a step S4 304 be per path 222 the sign messages (ie the traffic sign records) of all crossings to a list or sequence of path clusters 315 aggregated. Each path cluster 315 This represents a sign candidate, typically with position, orientation, type, various additional information (eg temporal restrictions of validity) and the amount of sign messages associated with it. With the help of a cluster algorithm, preferably a customized variant of the K-Means algorithm, can be used for a path 222 the constellation or sequence of path clusters 315 which are the ones for the path 222 Existing sign messages (ie traffic sign records) best explained. As a measure of the quality of the solution found, the log likelihood of the constellation of the path p with respect to the given sign messages can be used (as an optimization criterion):

With

where n _crossings the number of relevant crossings (of the path p) and ni / sign messages corresponds to the number of sign messages available for crossing i, and where δ _ij = 1 if the sign message j of the crossing i is a path cluster 315 was assigned (otherwise δ _ij = 0). LLH ij / assignment corresponds to the log likelihood of the sign message j of crossing i given the position of its associated path cluster or an empirically determined value depending on the driving situation, if no path cluster has been assigned. With the "position" of a sign 104 , a path cluster 315 or a supercluster 316 may be meant its actual position and possibly its orientation. The procedure described in this document 300 is applicable accordingly if the respective path cluster 315 , Supercluster 316 or shield 104 is modeled without orientation.

p ₁ corresponds to the probability that any fixed sign message (ie a traffic sign record) is a hoax, np / path cluster corresponds to the number M for the path p considered path clusters 315 , ε _ik = 1 if the path cluster k has been assigned a sign message of the crossing i (otherwise ε _ik = 0), p ₂ corresponds to the probability that one on the path 222 actual sign 104 when driving through the path 222 from the camera system 102 a measuring vehicle 101 is detected and reported and n _Prior corresponds to a fixed number of virtual crossings where no signs 104 were reported.

By considering the probability p ₂ for detecting a path traveled on the path 222 located sign 104 At the same time, the probability of the counter-event, ie the non-detection of an actual shield, can also be determined 104 , be determined. So can the actual number of existing signs 104 on a path 222 be determined with increased accuracy. In particular, as the number of passes increases, the growing number of sign messages (ie traffic sign records) does not result in the growing number of sign messages being transmitted through a larger number of path clusters 315 is explained as unnecessary path clusters 315 to the same extent, ie proportional to the number of crossings, by the comparatively small log likelihood of non-detection of an existing sign 104 by passing through the associated path 222 "to get punished.

The calculation of LLH ij / assignment is done using a probabilistic model that estimates the likelihood of issuing a particular sign message (ie the detection of a traffic sign 104 ) depending on the actual position and orientation of the sign 104 , its type and additional information and / or the position, orientation, speed and / or rotation rate of the vehicle 101 describes. Advantageously, for the modeling of LLH ij / assignment Likelihood models normal distributions are used, which are parameterized on the basis of at least one of the above sizes. Thus, in particular the influence of speed and / or rate of rotation of the vehicle 101 the accuracy of the reported shield positions should be taken into account so as to avoid the identification of additional false shield positions, even in the case of a large spread of reported shield positions, in order to achieve a higher reliability of the result.

A possible procedure for determining the path clusters 315 using a fitted variant of the K-Means algorithm is in 4 shown. In step S4a 401 becomes the number of clusters to be discovered 315 a path 222 initially set to 0 and the resulting value of LLH p / path determined. Following is the number 412 the cluster 315 increases to 1 and it goes to step S4b 402 continued. In this step 402 become the clusters 315 (their number 412 in step 401 was established) taking into account the existing shield messages (ie traffic signs records) suitably initialized, so that initial or output cluster positions 413 to be provided. In step S4c 403 the assignment of shield messages to clusters occurs, so that the value of the log likelihood LLH p / path is maximized. This takes into account that the camera 102 of a vehicle 101 (except for false reports) per real existing sign 104 only a maximum of one message is output, so that different messages the same crossing of different signs 104 must come from. This condition can be taken into account in such a way that only a maximum of one sign message of the same crossing can be assigned to each cluster. At the same time, a sign message of a crossing is only assigned to a maximum of one cluster. Advantageously, the resulting assignment problem can be solved using the Hungarian method or the Kuhn-Munkres algorithm. This results in an optimal allocation 414 of traffic sign records of different crossings to different clusters. The procedure described makes it possible to take into account the fact that the camera 102 of a vehicle 101 (except for false reports) per real existing sign 104 only a maximum of one message is output, so that different messages the same crossing of different signs 104 and thus the minimum number of existing signs 104 can be better determined.

Based on the optimal allocation determined in this way 414 The sign messages for the clusters will then be the optimal position 415 the cluster (step S4d 404 ). When using normal distributions to model LLH ij / assignment can be the optimal position 415 of a cluster can be determined efficiently analytically. Steps S4c, 403 and S4d, 404 are repeated until either the maximum number of iterations (eg 10) has been reached (abort condition 405 ) or until the position 415 the cluster has not changed significantly compared to the last run (abort condition 406 ). For every number 412 of clusters (except 0), steps S4b 402 , S4c 403 and S4d 404 with different inital cluster positions 413 be repeated (eg 10 times) to avoid local minima.

After completion of a complete run of steps S4a, S4b, S4c and S4d, it is checked whether the best value of LLH p / path of all cluster constellations examined in it the best value of LLH p / path the previous (smaller) number 412 the cluster (0 in the first pass, 1 in the second pass, etc.) surpasses. If this is the case, the corresponding value and the associated cluster constellation are stored as a preliminary result and the number 412 the cluster in S4a 401 for a new run of steps S4b 402 , S4c 403 and S4d 404 increased by one (termination condition 407 ). Is the best value of LLH p / path (ie the optimization criterion) of the current number 412 of clusters worse than the previous number 412 of clusters, the cluster constellation belonging to the previous number of clusters is output as the final result of step S4 (abort condition 407 ). Thus, in an iterative way, the optimum number M 412 of path clusters 315 and the optimal cluster positions 415 the M path cluster 315 be determined.

A variant of step S4 304 consists of sorting the existing sign messages (ie traffic sign records) in advance by type and / or additional information and the path clusters 315 determined separately for each of the resulting groups (eg for each traffic sign type) according to the procedure described above. The calculation of the path clusters 315 This makes it more efficient. Furthermore, the accuracy of the path clusters can be determined 315 and cluster positions 415 increase.

Another variant of step S4 304 is in place of LLH ij / assignment Use the square of the Mahalanobis distance between observed and actual shield position multiplied by minus 0.5. In the case of non-assignment of the sign message j pertaining to crossing i, LLH ij / assignment in this case be replaced by a heuristically determined constant, for example -4.5. This procedure is the same as the original procedure using normal distributions to model LLH ij / assignment but at a reduced computational cost.

As a result of step S4 304 you get for different paths 222 different sets of path clusters 315 , where a set of path clusters 315 the positions 415 of traffic signs 104 on a path 222 displays. The to a first path 222 associated set of path clusters 315 represents the signs 104 that from this first path 222 out through the sensors 102 a measuring vehicle 101 can be detected. However, any of these signs can 104 potentially also from another second path 222 be seen from - either because part of this second path 222 with a part of the first path 222 is identical, or because the sign in question 104 for example, from a road section 221 the second path 222 is visible from the parallel to a road section 221 of the first path 222 runs. In step S5 305 can therefore be the path clusters 315 the different paths 222 to so-called superclusters 316 be merged to overlap between the sets of path clusters 315 different paths 222 to detect and thus the accuracy of the determined cluster positions 415 continue to increase.

Every supercluster 316 again represents a shield candidate, z. B. with position, orientation, type and / or various additional information (eg. Time restrictions on validity). Supercluster 316 can be similar to path clusters 315 be determined. In this case, instead of individual sign messages (ie individual traffic sign records) the superclusters 316 individual path clusters 315 assigned. As it is with the determined path clusters 315 a path 222 around different signs 104 is allowed to each supercluster 316 typically only a maximum of one path cluster 315 from a path 222 be assigned.

In addition to a supercluster 316 associated path clusters 315 assigns each supercluster 316 typically the information on which discrete road section 221 and to what direction the supercluster 316 (ie that through the supercluster 316 displayed sign 104 ) heard on which street side the sign 104 stands or whether the shield 104 z. B. is attached to a gantridge over the road. Such a discrete street section 221 can in principle have any but fixed length, which, however, may not exceed a predetermined maximum length of, for example, 50 meters. The definition of road sections 221 can once before performing step S5 305 take place (as stated above).

mit

wobei n_Pfade der Zahl der betrachteten Pfade 222 und n i / Pfadcluster der Zahl der zu Pfad i vorhandenen Pfadcluster 315 entspricht. δ_ij = 1 falls das Pfadcluster j des Pfades i einem Supercluster 316 zugeordnet wurde (sonst δ_ij = 0). LLH ij / Zuordnung entspricht der Log-Likelihood des Pfadclusters j von Pfad i gegeben der Position des ihm zugeordneten Superclusters 316 bzw. einem empirisch ermittelten Wert, falls kein Supercluster 316 zugewiesen wurde. p ij / 1 entspricht der Wahrscheinlichkeit, dass es sich bei den zu Pfadcluster j des Pfades i gehörenden Schildmeldungen um Falschmeldungen handelt, wobei vorteilhafterweise auch für alle Pfade i und Pfadcluster j der gleiche Wert für p ij / 1 gewählt werden kann. n_Supercluster entspricht der Zahl der betrachteten Supercluster 316. ε_ik = 1 falls dem Supercluster k ein Pfadcluster 315 des Pfades i zugeordnet wurde (sonst ε_ik = 0). p ik / 2 entspricht der empirisch ermittelten Wahrscheinlichkeit, dass das Supercluster k von Pfad i aus sichtbar ist und gemeldet wird, wobei unterschieden wird, ob der dem Supercluster 316 zugeordnete Straßenabschnitt 221 Teil von Pfad i ist oder nicht. LLH k / Clusterposition entspricht der Log-Likelihood der Position der Supercluster 316, gegeben der gewählten Zuordnung zu Straßenabschnitten 221, Fahrtrichtungen und/oder Straßenseiten. LLH_{Konfiguration} entspricht der Log-Likelihood eben dieser Zuordnung, wobei LLH_{Konfiguration} insbesondere auch Erfahrungswerte berücksichtigen kann, bspw. dass Schilder auf Autobahnen häufig auf beiden Seiten der Fahrbahn platziert sind oder dass Geschwindigkeitsbegrenzungsschilder in der Regel nicht direkt vor sondern eher nach Kreuzungen anzutreffen sind.For a given number of superclusters 316 with assigned road sections 221 , Directions and / or road sides (including the possibility of belonging to a gantry) should be the most plausible assignment of path clusters 315 , the most plausible position of the supercluster 316 and the position-dependent measure LLH _Global for this plausibility (as a supercluster optimization criterion). The measure LLH _Global for the plausibility of such an assignment is, for example, from the following equations:

With

where n _{paths are} the number of paths considered 222 and ni / path cluster the number of path clusters existing for path i 315 equivalent. δ _ij = 1 if the path cluster j of the path i is a supercluster 316 was assigned (otherwise δ _ij = 0). LLH ij / assignment corresponds to the log likelihood of path cluster j of path i given the position of its associated supercluster 316 or an empirically determined value, if no supercluster 316 was assigned. p ij / 1 corresponds to the probability that the sign messages belonging to path clusters j of path i are false reports, and advantageously also the same value for all paths i and path cluster j p ij / 1 can be chosen. n _Supercluster corresponds to the number of considered superclusters 316 , ε _ik = 1 if the supercluster k is a path cluster 315 of the path i was assigned (otherwise ε _ik = 0). p ik / 2 corresponds to the empirically determined probability that the supercluster k is visible from path i and reported, distinguishing whether the supercluster 316 assigned road section 221 Part of Path i is or not. LLH k / cluster position The log likelihood corresponds to the position of the supercluster 316 given the chosen assignment to road sections 221 , Directions and / or street sides. LLH _{configuration} corresponds to the log likelihood of just this assignment, with LLH _{configuration} in particular as well Experience can take into account, for example, that signs are often placed on highways on both sides of the road or that speed limit signs are usually not found directly before but rather after intersections.

By a suitable choice of LLH ij / assignment . p ij / 1 and p ik / 2 In addition to the measurement errors resulting from the individual sign messages (non-recognition and incorrect recognition as well as deviations in the reported position), systematically occurring registration errors can also be taken into account. This allows the consideration of situations in which, for example, an existing sign 104 from a particular path 222 is not visible or in which the detection is always subject to a specific error. The explicit consideration of such situations allows a more reliable fusion of the existing information and a more accurate determination of the actual shield position. Thus, it is possible to improve the accuracy of the detected shield positions by taking into account systematic errors in the reported shield positions (ie in the position data of the traffic sign data sets) in a probabilistic model and thereby less affect the result.

A possible procedure for determining the supercluster 316 using a fitted variant of the K-Means algorithm is in 5 shown. In step S5a 501 will first be a number of supercluster candidates 512 determined by adding for each road section 221 , each direction of travel and / or each street side first a supercluster 315 (For example, on the middle of the road section 221 with orientation according to the direction of travel) is instantiated. It can then be the K-Means algorithm, as it later in steps S5c 514 and S5d 515 is used with this supercluster 316 is performed as the only selected cluster candidate. The resulting log likelihood is compared with the LLH in the event that no cluster candidate is selected at all. If the higher LLH results when using the supercluster, this comes as a supercluster candidate 316 for the actual algorithm in question.

A special feature is the street side, which represents gantries. Gantries usually have multiple signs, while a supercluster 316 typically represents exactly one sign. In a special feature of the algorithm are therefore for the road side "sign bridge" for each road section 221 and different numbers of super clusters for each direction of travel 316 to investigate for the determination of the shield candidate. The number of superclusters 316 with the highest LLH is subsequently considered a supercluster candidate 512 included, ie the supercluster candidate 512 In the case of "gantries" there are several signs 104 ,

Each of the supercluster candidates determined this way 512 has the property that its use the existing path clusters 315 of different paths 222 better explained than the use of no supercluster candidate 512 , However, it may be that the LLH (Log-Likelihood) for the simultaneous existence of the supercluster of two candidates is smaller than that of the existence of only one of the two superclusters. So the goal is to find the set of shield candidates where the coexistence of the associated supercluster best explains the existing path clusters or maximizes LLH _Global .

Based on the identified supercluster candidates 512 Therefore, in step S5b 502 one subset each 513 the candidate 512 selected to match the subset of candidates 512 To determine the value of LLH _Global resulting from the concurrent existence of the associated supercluster. The choice of candidates depends on the method used to solve the problem of discrete optimization. A complete search through the space of the possible subsets, a variant of this with premature termination under certain conditions ("branch-and-bound") or even local methods such as local-hill-climbing would be conceivable.

When using local hill climbing methods, the value of LLH _{Global is} incrementally increased by a fraction of the last accepted quantity 513 of candidates 512 only a small change, for example the addition or removal of one or more candidates, is allowed. The amount modified in this way 513 is exactly then as a new amount 513 of candidates 512 accepted if the resulting value of LLH _{Global is} greater than the previous amount 513 is. The algorithm ends when there is no acceptable change that would increase the value of LLH _Global .

Advantageously, for step S5b 502 used such a local hill climbing method. To avoid local optima, local changes are the simultaneous addition and removal of a maximum of n candidates 512 , preferably n = 2, admitted.

To check the consistency of the results with those already working on the neighboring cells 212 for each of these existing shield positions, including their assignment to road sections 221 , Directions and / or road sides also a supercluster 316 created. These already existing, "old" superclusters 316 be both in the identification of candidates 512 for the "new" supercluster 316 but also in the search for their optimal combination, but are not allowed in their position or assignment to road sections 221 , Directions of travel and / or street sides are changed.

The determination of too much 513 of candidates 512 The associated value of LLH _Global is largely analogous to steps S4c 413 and S4d 414 , Starting from the initial positions of the old and new superclusters, in step S5c 503 First, the optimal allocation 514 the path cluster 315 to the old and new superclusters, while in step S5d 514 the position 515 excluding the new supercluster, taking into account the associated path clusters 315 is optimized. In both cases, LLH _Global serves as a measure of the quality of the solution found. As mentioned above, every supercluster 316 in this case per path 222 only one path cluster 315 be assigned. The steps S5c 514 and S5d 515 are repeated until either a maximum number of iterations has been reached (abort condition 505 ) or no change in the cluster positions 515 yields more (termination condition 506 ).

The adjustment of the amount 513 the supercluster candidates to use 512 in step S5b 502 and the subsequent calculation of LLH _Global in steps S5c 503 and S5d 503 are iterated until there is no allowable change in the amount 513 the candidate to use 512 There is more that would make LLH _Global even better. The amount accepted to the last one 513 of candidates 512 belonging to "new" supercluster 316 thus provide the result of step S5 305 dar. The optimal positions of the supercluster 316 arise as the cluster positions 515 the quantity accepted to the last 513 of candidates 512 belonging to "new" supercluster 316 ,

Under certain circumstances, the determined "new" supercluster positions 515 have systematic deviations from the actual shield positions. The probability distribution for such deviations became the one in S5 305 used LLH k / cluster position for each of the supercluster k already considered. As a result, the superclusters represent 316 each the position of a shield 104 including the systematic deviation.

mit LLH k / Syst.Abweichung der Log-Likelihood der systematischen Abweichung des Superclusters k von der Position des zugehörigen Schildes, und LLH k / Schildposition der Log-Likelihood der Position des zugehörigen Schildes und LLH_{Konfiguration} der bekannten Log-Likelihood der Zuordnung der Supercluster zu Straßenabschnitten 221, Fahrtrichtungen und/oder Straßenseiten, die durch ein geeignetes Verfahren zur Variation der Schildpositionen maximiert wird.Starting from the position 515 a supercluster 316 can in a step S6 306 the most likely position 317 of the shield 104 determined by

With LLH k / syst. Deviation the log likelihood of the systematic deviation of the supercluster k from the position of the associated shield, and LLH k / shield position the log likelihood of the position of the associated shield and LLH _{configuration of} the known log likelihood of the assignment of the supercluster to road sections 221 , Directions of travel and / or road sides maximized by a suitable method of varying shield positions.

As described above, the calculation of the shield positions 317 of a larger area, by dividing this area into many small subregions, so-called cells 211 , is decomposed. For each of these cells 211 There is an inner and an outer bounding box. Basically, only those are in step 56 306 determined shield positions (ie cluster positions) 317 stored in the inner bounding box 211 As the other shield positions are later during the processing of the adjacent cells 212 be determined once again if they are in the inner bounding box of each cell. It can thus the determined shield positions 317 outside the inner bounding box 211 to be discarded (step S7 307 ).

In a special form of the procedure 300 In addition, those newly determined shield positions 317 that are in a certain area of the inner bounding box of adjacent and already processed cells, added to the shield positions of the respective cells. This avoids that certain signs are not stored at all, since they were outside the inner bounding box when processing the neighboring cells, but in the processing of the current cell due to the changed amount of input data in the inner bounding box of the adjacent Cell were located. As a result, as well as by the steps S1 301 and S2 302 described procedure of dividing a region into individual cells 211 , as well as the consideration of already determined shield positions in step S5 305 , it becomes possible the procedure 300 to execute in a computationally efficient manner on arbitrarily large areas.

The present invention is not limited to the embodiments shown. In particular, it should be noted that the description and figures are intended to illustrate only the principle of the proposed methods, apparatus and systems.

Claims (18)

Procedure ( 300 ) for determining positions of traffic signs ( 104 ) in a road network ( 200 ), the process ( 300 ), - determining ( 303 ) of a path ( 222 ) by a subarea ( 210 ) of the road network ( 200 ); - Determine ( 303 ) of measurement data of a plurality of journeys of one or more vehicles ( 101 ) along the path ( 222 ); where the measured data are determined by means of environment sensors ( 102 ) of one or more vehicles ( 101 ) were recorded; wherein the measurement data of a journey comprises one or more traffic sign data sets for corresponding one or more detected traffic signs ( 104 ) when driving along the path ( 222 ) were detected; wherein a traffic sign record for a detected traffic sign ( 104 ) Position data regarding a position of the detected traffic sign ( 104 ); and - determining ( 304 ), based on the measured data, a number M ( 412 ) of clusters ( 315 . 316 ) for grouping the traffic sign records and positions ( 415 . 317 ) the M cluster ( 315 . 316 ) along the path ( 222 ), so that an optimization criterion is improved; where a cluster ( 315 . 316 ) at least one traffic sign ( 104 ) indicates; where the position ( 415 . 317 ) of a cluster ( 315 . 316 ) the position of at least one traffic sign ( 104 ) indicates; and wherein the optimization criterion depends on a probability that a traffic sign ( 104 ) while driving along the path ( 222 ) from the environment sensor system ( 102 ) of one or more vehicles ( 101 ) is detected. Procedure ( 300 ) according to claim 1, wherein - the path ( 222 ) actual traffic signs ( 104 ) at actual positions represented by the measurement data; and the optimization criterion indicates the probability that the determined positions ( 415 . 317 ) the M cluster ( 315 . 316 ) the actual positions of the actual traffic signs ( 104 ) along the path ( 222 ) correspond. Procedure ( 300 ) according to one of the preceding claims, wherein the optimization criterion of a probabilistic model for measurement errors in the position data with respect to the position of the detected traffic sign ( 104 ) depends. Procedure ( 300 ) according to one of the preceding claims, wherein the optimization criterion depends on a probability that a detected traffic sign ( 104 ) while driving along the path ( 222 ) is a misdetection. Procedure ( 300 ) according to one of the preceding claims, wherein - the optimization criterion depends on a probability model which determines the probability of detection of a traffic sign ( 104 ) depending on a condition of a vehicle ( 101 ) while driving along the path ( 222 ) indicates; and - the condition of the vehicle ( 101 ) in particular of a speed and / or a rotation rate of the vehicle ( 101 ) depends. Procedure ( 300 ) according to any one of the preceding claims, wherein the determining ( 304 ) of the positions ( 415 . 317 ) the M cluster ( 315 . 316 ), Assign ( 403 ) of the traffic sign records of the plurality of trips to the M clusters ( 315 . 316 ) such that - a cluster ( 315 . 316 ) is assigned a maximum of only one traffic sign record of a trip of the plurality of trips; and - a traffic sign record only one cluster ( 315 . 316 ). Procedure ( 300 ) according to claim 6, wherein the determining ( 304 ) of the positions ( 415 . 317 ) the M cluster ( 315 . 316 ), determining ( 404 ) of the position ( 317 ) of a cluster ( 315 . 316 based on the position data of the one or more traffic sign records associated with the cluster ( 315 . 316 ). Procedure ( 300 ) according to one of claims 6 to 7, wherein the optimization criterion of a distance between the determined position ( 317 ) of a cluster ( 315 . 316 ) and positions of one or more detected traffic signs ( 104 ) whose one or more traffic sign records belong to the cluster ( 315 . 316 ) assigned. Procedure ( 300 ) according to any one of the preceding claims, wherein the determining ( 304 ) the number M ( 412 ) of clusters ( 315 . 316 ), - determining ( 404 ) of a value of the optimization criterion for different numbers ( 412 ) of clusters ( 315 . 316 ); and - determining ( 304 ) the number M ( 412 ) based on the values of the optimization criterion for the different numbers ( 412 ) of clusters ( 315 . 316 ). Procedure ( 300 ) according to one of the preceding claims, wherein the optimization criterion for a path p comprises

With

wherein - n corresponds to the number of _crossings rides along the path p; - ni / sign messages the number of traffic signs records for the i ^th driving corresponds; - LLH ij / assignment the likelihood, in particular the log likelihood, indicates that the traffic sign data set j of the i ^th drive belongs to a cluster ( 315 . 316 ) assigned; - LLH ij / assignment from a distance between a position of the detected traffic sign ( 104 ) of the traffic sign data set j of the i ^th drive and the determined position ( 317 ) of a cluster ( 315 . 316 ), to which the traffic sign data set j of the i ^th drive is assigned, or corresponds to an empirically determined value, if the traffic sign data set j of the i ^th drive does not correspond to a cluster ( 315 . 316 ) assigned; - δ _{ij takes} the value 1 or 0, depending on whether the traffic sign data record j of the i ^th drive is a cluster ( 315 . 316 ) or not; P ₁ indicates a probability that a traffic sign record is a hoax; - np / path cluster the number of clusters ( 315 . 316 ) on the path p; - ε _ik assumes the value 1 or 0, depending on whether or not the cluster k has been assigned a traffic sign data set for the ^ith trip; P ₂ indicates a probability that an actual traffic sign (p) located on the path p ( 104 ) is detected when driving along the path p; and - n _prior corresponds to a number of virtual trips where no traffic signs ( 104 ) are detected. Procedure ( 300 ) according to any one of the preceding claims, wherein the process ( 300 ), - determining ( 304 ) of the positions ( 415 ) of clusters ( 315 ), hereinafter referred to as path cluster ( 315 ), for a first path ( 222 ) and for an at least partially different second path ( 222 ); - the first path ( 222 ) a first road section ( 221 ) and the second path ( 222 ) a second road section ( 221 ) of the road network ( 200 ); - the first section of road ( 221 ) and the second road section ( 221 ) are designed such that when driving along the second road section ( 221 ) a traffic sign ( 104 ) on the first section of road ( 221 ) can be detected; - Determine ( 305 ), based on the path clusters ( 315 ) for the first and second paths ( 222 ), one position ( 317 ) of a supercluster ( 316 ) for the first section of road ( 221 ), so that a supercluster optimization criterion is improved; where the supercluster ( 316 ) at least one traffic sign ( 104 ) indicates; and where the position ( 317 ) of the supercluster ( 316 ) the position of the at least one traffic sign ( 104 ). Procedure ( 300 ) according to claim 11, wherein the supercluster optimization criterion depends on a probability that a traffic sign ( 104 ) of a road section ( 221 ) while driving along another road section ( 221 ) can be detected. Procedure ( 300 ) according to any one of claims 11 to 12, wherein the supercluster optimization criterion of a plausibility of the determined position ( 317 ) of the supercluster ( 316 ) for the first section of road ( 221 ) with respect to typical positions of traffic signs ( 104 ) in a road section ( 221 ) depends. Procedure ( 300 ) according to one of claims 11 to 13, wherein the supercluster optimization criterion is based on a distance between the determined position ( 317 ) of the supercluster ( 316 ) and the positions of the one or more path clusters ( 315 ) depends on the supercluster ( 316 ) assigned. Procedure ( 300 ) according to one of claims 11 to 14, wherein the supercluster optimization criterion comprises

With

where n _{paths of} the number of paths ( 222 ) required for the identification of superclusters ( 316 ) to be viewed as; - ni / path cluster the number of path clusters ( 315 ) of the i ^th path corresponds; - LLH ij / assignment, supercluster the probability, in particular the log likelihood, indicates that the path cluster j of the i ^th path is a supercluster ( 316 ) assigned; - LLH ij / assignment, supercluster from a distance between a position of the path cluster j of the i ^th path and the determined position ( 317 ) of a supercluster ( 316 ) Depends on the path of the cluster j of the i ^th path is associated with or corresponds to an empirically determined value, if the path cluster j of the i ^th path (no supercluster 316 ) assigned; - δ ^S _{ij takes} the value 1 or 0, depending on whether the path cluster j of the i ^th path is a supercluster ( 316 ) or not; - p ij / 1 indicates a probability that the one or more traffic sign records associated with the path cluster j of the ^ith path are false reports; - n _{supercluster of} the number of superclusters to be determined ( 316 ) corresponds; - ε ^S _{ik takes} the value 1 or 0, depending on whether the supercluster k has been assigned a path cluster of the i ^th path or not; - p ik / 2 indicates a probability that a traffic sign ( 104 ) of the supercluster k is visible from the i ^th path and detected; - LLH k / cluster position a probability, in particular a log likelihood, of the position determined ( 317 ) of a supercluster ( 316 ) for the road section ( 221 ), to which the supercluster ( 316 ) has been assigned; and LLH _{configuration} a probability, in particular a log likelihood, of the association between supercluster ( 316 ) and road section ( 221 ). Procedure ( 300 ) according to any one of claims 11 to 15, wherein - the method ( 300 ), determining the positions ( 415 ) of M _l path clusters ( 315 ) for L different paths ( 222 ); - L is an integer, with L>0; M _{l is} the number of path clusters ( 315 ) on the ^long path ( 222 ), with l = 1, ..., L; and - the method ( 300 ), determining, based on the path clusters ( 315 ) the L different paths ( 222 ), a number N of superclusters ( 316 ), so that the supercluster optimization criterion is improved. Procedure ( 300 ) according to one of the preceding claims, wherein - the subarea ( 210 ) a first cell ( 211 ) through a second cell ( 212 ) is at least partially enclosed; - the procedure ( 300 ), Discard ( 307 ) of a cluster ( 315 . 316 ) the M cluster ( 315 . 316 ) whose position in the second cell ( 212 ) and / or not in the first cell ( 211 ) lies; and - the method ( 300 ) for a further subarea ( 210 ), in which the second cell ( 212 ) by one or more additional cells ( 212 ) is at least partially enclosed. Evaluation unit for determining positions of traffic signs ( 104 ) in a road network ( 200 ), wherein the evaluation unit is set up, - a path ( 222 ) by a subarea ( 210 ) of the road network ( 200 ) to investigate; - Measurement data of a plurality of journeys of one or more vehicles ( 101 ) along the path ( 222 ) to investigate; where the measured data are determined by means of environment sensors ( 102 ) of one or more vehicles ( 101 ) were recorded; wherein the measurement data of a journey comprises one or more traffic sign data sets for corresponding one or more detected traffic signs ( 104 ) when driving along the path ( 222 ) were detected; wherein a traffic sign record for a detected traffic sign ( 104 ) Position data regarding a position of the detected traffic sign ( 104 ); and, on the basis of the measured data, a number M ( 412 ) of clusters ( 315 . 316 ) for grouping the traffic sign records and positions ( 415 . 317 ) the M cluster ( 315 . 316 ) along the path ( 222 ), so that an optimization criterion is improved; where a cluster ( 315 . 316 ) at least one traffic sign ( 104 ) indicates; where the position ( 415 . 317 ) of a cluster ( 315 . 316 ) the position of at least one traffic sign ( 104 ) indicates; and wherein the optimization criterion depends on a probability that a traffic sign ( 104 ) while driving along the path ( 222 ) from the environment sensor system ( 102 ) of one or more vehicles ( 101 ) is detected.

Images