METHOD, DEVICE AND COMPUTER PROGRAM FOR PROCESSING DATA ABOUT A TRAFFIC LIGHT SYSTEM

Description

[0001] The present invention relates to a method, a device and a computer program for processing data about a traffic light system.

[0002] There are automatic driving functions of a motor vehicle that process information from traffic lights and surrounding vehicles in order to react accordingly in traffic.

[0003] DE 10 2013 2019 206 A1 discloses the processing of historical data representative of past signal programs of a traffic signal system. From this historical data, models in the form of state machines can be derived, which can be used to predict the future state of a signal group of the traffic signal system.

[0004] DE 10 2017 127 346 A1 describes a system, a method, and a device for detecting brake lights. Using data from a distance sensor or camera image data, a zone within an image frame corresponding to a vehicle is detected. Deep neural networks configured for brake lights classify a vehicle's brake light as on or off based on image data in the zone.

[0005] US 2018 / 0285664 A1 describes a method for identifying a signal state of a traffic light system. A camera captures an image of traffic lights, which is then displayed, for example, on a touchscreen. A user selects a traffic light to be monitored. The signal state of the selected traffic light is recorded. As soon as the selected traffic light turns green, a corresponding signal is output. Furthermore, the traffic light can also be selected automatically by using a result from a lane keeping or lane change assistant as input for a traffic light recognition system.

[0006] DE 10 2018 007 962 A1 describes a method for detecting traffic light positions. A plurality of fleet vehicles capture a traffic scene at a traffic light intersection using a camera, and traffic light candidates are determined in the captured traffic scene and uploaded to a central backend. In the respective fleet vehicle, the geometric positions of the traffic light candidates relative to each other are determined and uploaded to the backend. If a plurality of uploaded geometric positions for the same traffic scene are present, a statistical analysis of this plurality of uploaded geometric positions is performed in the backend, such that frequently occurring traffic light candidates are interpreted as traffic lights, while rarely occurring traffic light candidates are interpreted as false detections.From the geometric positions interpreted as traffic lights, a pattern with confidence intervals of the traffic light positions is created and assigned to the respective traffic light intersection and a specific direction of travel.

[0007] The challenge, therefore, is that the information relevant for an automatic vehicle function is determined not by the driver, but by a method, device, and / or computer program. Particularly for a vehicle traveling on a section of road controlled by a traffic light system and approaching the traffic light system, the relevant traffic light information, such as the position of the traffic light on the vehicle's path, must be reliably recorded. This relevant traffic light information can then also be used as input for other driving functions.

[0008] The object of the present invention is to provide a method, a device and a computer program that at least partially overcome the above-mentioned challenge.

[0009] This object is achieved by a method according to claim 1, a device according to claim 11 and a computer program according to claim 12.

[0010] Further advantageous embodiments of the invention emerge from the subclaims and the following description of preferred embodiments of the present invention.

[0011] A first aspect of the disclosure relates to a method for processing data about a traffic light system controlling a road section with at least one traffic light. The method comprises the following steps: Retrieving traffic light objects that represent traffic light information about the traffic light system; detecting at least one cluster in the traffic light objects, wherein a detected cluster represents a corresponding (traffic light) of the at least one traffic light of the traffic light system; assigning the traffic light objects to the at least one cluster; and deriving a spatial position of the at least one traffic light based on the assigned traffic light objects

[0012] Traffic light information about a traffic light system is recorded by a detection device, which is attached to a vehicle, for example. Using a swarm of vehicles equipped with detection devices, a multitude of traffic light information items can be recorded about the same traffic light system. This traffic light information is represented by traffic light objects. The traffic light objects correspond to data or data sets. The traffic light objects can be stored in a memory / database. The term "traffic light" refers to a physical traffic light. A traffic light system comprises at least one traffic light.

[0013] The traffic light information about the traffic light system includes position data of (at least one) traffic light in the traffic light system. In addition, the traffic light phase sequences can also be recorded as traffic light information.

[0014] The position of the detected traffic light is recorded relative to the vehicle located on the roadway section and moving toward the traffic light. The position, and thus the movement of the vehicle, can be determined, for example, using a vehicle location determination system, such as a GPS system. This allows the position of the vehicle itself to be represented in a global coordinate system. Thus, the traffic light position recorded relative to the vehicle can be converted into an absolute traffic light position in this global coordinate system. Alternatively or in addition to the global coordinate system, a format used in GPS systems or a location used, other x, y, z coordinate sets, or other suitable location information can be used.

[0015] In particular, a traffic light object can include data about the position history of the detected traffic light. The position relative to a vehicle can be detected using the detection device. The position history can be represented by a plurality of time-position pairs.

[0016] One step involves detecting at least one cluster among the traffic light objects. This detects and groups traffic light objects with similar properties. In this case, the traffic light objects are clustered according to their traffic light position data. Each detected cluster (group) represents a corresponding traffic light in the traffic light system.

[0017] The traffic light objects are assigned to the detected clusters. In other words, the traffic light objects are provided with additional information so that it is clear in which cluster they are located.

[0018] The spatial position of at least one traffic light can be derived using the associated traffic light objects. In other words, the spatial position of the traffic light represented by the identified cluster can be derived from a detected cluster. Thus, the traffic light objects assigned to the corresponding identified cluster are used to determine the spatial position of a traffic light in a traffic light system using the position data they contain. The spatial position is the actual position of the (physical) traffic light and can be represented in the global coordinate system.

[0019] Thus, the traffic light objects can be used to create a virtual image of the traffic light system, a so-called traffic light image. This traffic light image can be saved, particularly in a database. The traffic light image thus represents the spatial position of at least one traffic light in the traffic light system in absolute coordinates within the global coordinate system. The traffic light image, which represents the traffic light system, can be used as input information for various vehicle functions.

[0020] According to the invention, retrieving the traffic light objects further comprises: Retrieving a course of the roadway section; retrieving a traffic light line indicating a position of the traffic light system along the roadway section; and discarding the traffic light objects whose distance from the traffic light line exceeds a predetermined limit distance.

[0021] The course of the road section can be recorded, for example, by the vehicle-side detection device mentioned above and, in particular, derived from the vehicle's road course. This road course can be represented by a plurality of time-position pairs, whereby the position of the vehicle in the global coordinate system is meant. With a plurality of vehicles, a plurality of road courses are recorded. The course of the road section can be derived from the average of the plurality of road courses. In particular, the course of the road section can correspond to a longitudinal axis of the road section. The direction of travel of the road section refers to the direction in which the vehicle is moving towards the traffic light system (which controls the road section).

[0022] In general, the traffic light line is the position along the roadway section at which a vehicle passes the traffic light. In the global coordinate system mentioned above, the traffic light line can be understood as a straight line perpendicular to the longitudinal axis of the roadway section.

[0023] The traffic light line can correspond to the position along the road section where, in the direction of travel along the road section, a traffic light object of the aforementioned traffic light objects is located for the first time. In other words, the traffic light line is located along the road section at the position where the vehicle first passes a traffic light object. In particular, the traffic light line can be determined specifically for each crossing (and thus also for each vehicle).

[0024] The distance of a traffic light object to the traffic light line is (from a bird's eye view of the roadway section) the shortest distance between the traffic light line and the traffic light object. In other words, the distance of a traffic light object to the traffic light line is the shortest distance between the traffic light object and a traffic light line plane, where the traffic light line lies in the traffic light line plane and the traffic light line plane is perpendicular to the longitudinal axis of the roadway section.

[0025] By discarding the traffic light objects according to the above condition, traffic light objects that are highly likely to represent pedestrian traffic lights and / or traffic lights for another roadway section can be filtered out. This "pre-filtering" allows traffic lights or traffic light objects for the roadway section controlled by the traffic light system to be more reliably identified.

[0026] Furthermore, the predetermined threshold distance can depend on a standard deviation from a median (value) of all distances of the traffic light objects to the traffic light line of the traffic light system. Instead of a fixed threshold, the predetermined threshold distance can thus depend on this standard deviation. This "pre-filtering" makes subsequent evaluation of the traffic light data independent of any scatter of the traffic light objects along the longitudinal direction of the roadway section (at the traffic light line). In particular, this "pre-filtering" eliminates the scatter.

[0027] According to the invention, the traffic light objects can be represented by a point cloud in the aforementioned coordinate system. In other words, the traffic light positions represented by the traffic light objects can be represented by corresponding coordinates (points) in the global coordinate system, so-called "traffic light object points." Thus, the traffic light objects can be particularly well (further) processed in the form of traffic light object points. In particular, mathematical operations can be applied to the traffic light objects.

[0028] In a variant, detecting the at least one cluster may further comprise: Projecting the point cloud onto a plane that perpendicularly intersects the roadway section at the traffic light line; and detecting the at least one cluster of traffic light objects using the projected point cloud.

[0029] The plane onto which the point cloud is projected corresponds to the traffic light line plane described above. The resulting projected points are clustered.

[0030] In an alternative, the detection of the at least one cluster in the traffic light objects may further comprise: Discard any cluster whose number of traffic light objects is less than a predetermined minimum number.

[0031] If the traffic light objects are represented by corresponding traffic light object points, any cluster whose number of traffic light object points is smaller than the predetermined minimum number is discarded.

[0032] If multiple clusters have been detected, the predetermined minimum number of traffic light objects can be dependent on the number of traffic light objects in the second-largest cluster. In particular, the predetermined minimum number can be one-fifth of the number of traffic light objects in the second-largest cluster. This ensures that clusters formed from scattered traffic light objects and thus do not correspond to a physical traffic light are filtered out.

[0033] Furthermore, the detection of the at least one cluster in the traffic light objects may include: Performing a principal component analysis on the at least one cluster; and if an eigenvalue of the at least one cluster exceeds a predetermined boundary eigenvalue: detecting at least one sub-cluster in the traffic light objects of the at least one cluster, wherein each detected sub-cluster represents a corresponding (traffic light) of the at least one traffic light of the traffic light system.

[0034] Principal component analysis is a multivariate statistical technique used to structure, simplify, and visualize large data sets by approximating a large number of statistical variables using a smaller number of meaningful linear combinations (principal components).

[0035] Principal component analysis is applied to each cluster detected in the traffic light objects. Specifically, principal component analysis is applied to the traffic light objects (traffic light object points) located in the respective clusters. This allows us to determine whether two closely spaced traffic lights (or the traffic light objects representing these two closely spaced traffic lights) are combined into a single cluster.

[0036] The principal components of the corresponding clusters can be calculated from the traffic light objects or traffic light object points, and the eigenvalues of the corresponding clusters can be determined. Clusters that exceed a predetermined threshold eigenvalue are identified as "double clusters," representing two closely spaced traffic lights. Using suitable methods / algorithms, e.g., a clustering method, subclusters can be identified from the traffic light objects belonging to such a double cluster. These subclusters of a double cluster represent a corresponding traffic light in the traffic light system. In addition, traffic light objects from a double cluster that cannot be assigned to a subcluster can be discarded.

[0037] By means of principal component analysis and the recognition of sub-clusters in the double clusters, traffic lights arranged close to each other can be reliably derived and distinguished

[0038] In an alternative, deriving the spatial position of the at least one traffic light may comprise: Forming a spatial average over the traffic light objects assigned to the cluster that represents the at least one traffic light.

[0039] The traffic light position data stored in the traffic light objects can therefore be used. The mean value of all traffic light position data of a cluster represents the spatial position of the traffic light represented by the corresponding cluster (within the global coordinate system). The totality of all spatial positions of all traffic lights in a traffic light system is the traffic light image for that traffic light system.

[0040] Alternatively, the at least one cluster can be identified using a first clustering method and / or the at least one subcluster can be identified using a second clustering method. Different clustering methods can be used. However, the same clustering methods can also be used, but with different method parameters.

[0041] As a rule, the well-known algorithm “Density-Based Spatial Clustering of Applications with Noise” (DBSCAN) is used. This algorithm is density-based and can detect multiple clusters. Noise points ( noise) are ignored and returned separately. The algorithm forms clusters of so-called "densely connected" points, i.e., points that are located no further than a specified distance (neighborhood length) from a "core point" in the same cluster. A core point is a point that is closer than the neighborhood length to at least a predetermined minimum number of other core points in the same cluster.

[0042] For example, the at least one cluster can be identified using a first clustering method, and the at least one subcluster can be identified using the DBSCAN algorithm. Different neighborhood lengths can then be used for the clustering methods.

[0043] Other clustering methods are also possible, e.g. appropriately trained (deep) neural networks.

[0044] In an alternative, at least one of the first and second clustering methods may comprise the DBSCAN algorithm.

[0045] Furthermore, a neighborhood length of the first clustering method can be larger than a neighborhood length of the second clustering method. This requires that the first and second clustering methods include the DBSCAN algorithm.

[0046] Furthermore, the traffic light objects are detected by a large number of vehicles.

[0047] According to the invention, the traffic light information is recorded by detection devices mounted on a large number of vehicles. By driving along the roadway and passing the traffic light system by a large number of vehicles, more information about the traffic light system can be collected. With each (consecutive) passing, a more precise analysis of the traffic light data can be performed.

[0048] Alternatively, the traffic light objects can be detected on a crossing-specific basis. "Crossing" refers to a vehicle from the multitude of vehicles moving on the section of road controlled by the traffic light system, approaching the traffic light system, and crossing it. During this crossing, the vehicle detects the traffic light system, particularly continuously, using its detection device.

[0049] A second aspect of the disclosure relates to a device for processing data from a traffic light system, wherein the device is designed and configured to carry out one of the methods described above.

[0050] A third aspect of the present disclosure relates to a computer program for processing data from a traffic light system, wherein the computer program is designed, when executed, to cause a data processing device to carry out one of the methods described above.

[0051] Embodiments of the invention will now be described by way of example and with reference to the accompanying drawings, in which: Fig. 1 schematically shows a road section controlled by a traffic light system; Fig. 2 schematically shows a method for processing data about the traffic light system; Fig. 3 schematically shows a traffic light image of the Figs. 1 Fig. 4 schematically shows an example of the traffic light system shown in Figs. 1 shown traffic light system; Fig. 5 schematically shows a method for grouping traffic lights of the Figs. 1 shown traffic light system; Fig. 6 schematically shows a method for determining a traffic light phase of a traffic light; and Fig. 7 schematically and exemplarily shows two phase curves.

[0052] In Figs. 11 schematically shows a road section Fb which is controlled by a traffic light system A. The road section Fb has three lanes F1, F2, F3, which in turn are controlled by four corresponding traffic lights tp 1 , tp 2 , tp 3 , tp 4 , tp 5 of the traffic light system A. The traffic light system A can be detected by a vehicle 1 traveling on the road section Fb. In particular, the individual traffic lights tp 1 - tp 5 of the traffic light system A are detected. For detection, the vehicle 1 has a detection device 3 which is designed to detect a phase curve and / or position of each traffic light tp 1 - tp 5. Furthermore, the detection device 3 can detect at least one of the position, position curve and orientation of the vehicle 1. The detection device 3 can, for example, comprise a camera device and a GPS device. The vehicle 1 has a radio interface 5 to communicate with a server / database (not shown) and exchange data.In the context of the disclosure, "detection by the vehicle 1" also means "detection by the vehicle-side detection device 3".

[0053] A traffic light's phase progression is also called a signal progression. A phase progression specifies phase durations and corresponding traffic light phases. A traffic light's phase progression can be represented as a multitude of time-traffic light phase pairs. A traffic light phase is a signal that a traffic light can display. Typically, a traffic light has the (traffic light) phases "green," "yellow," "red," and "yellow and red," with the traffic light usually displaying the phases in this order. Other sequences are also possible.

[0054] It is understood that vehicle 1 can also represent a plurality of vehicles, so that traffic light system A can be detected by a plurality of vehicles. Thus, for each crossing of traffic light system A by a vehicle 1 from the plurality of vehicles, a vehicle-specific, or otherwise referred to as a crossing-specific, data set can be created. This crossing-specific data set will be described later.

[0055] In the present traffic light system A, traffic lights tp 1 and tp 2 control lane F1, and traffic lights tp 3 and tp 4 control lanes F2 and F3. Thus, traffic lights tp 1 and tp 2 form a first traffic light group, and traffic lights tp 3 and tp 4 form a second traffic light group. Traffic lights tp 1 and tp 2 of the first traffic light group therefore have the same first phase progression, and traffic lights tp 3 and tp 4 of the second traffic light groups have the same second phase progression. In other words, the respective traffic lights of the traffic light groups have the same phase progression. Finally, lane F3 can also be controlled by a separate right-turn traffic light tp 5.

[0056] In the Figure 2 A method for creating a traffic light image 7 is shown. Traffic light image 7 is a data set that can be stored in a database, for example, and contains the positions of traffic lights in a traffic light system, for example the one in Figure 1shown traffic light system A with the traffic lights tp 1 - tp 5. In other words, the position of a traffic light system, and in particular its traffic lights, can be shown / represented by means of a traffic light image.

[0057] In a step S101, traffic light information, so-called traffic light objects, nc retrieved, e.g., from an existing database. Each traffic light object nc is representative of a corresponding traffic light of the traffic lights tp 1 -tp 5 and therefore includes at least position data of the corresponding traffic light. Furthermore, the traffic light objects include nc also information about the traffic light phases of the corresponding traffic lights.

[0058] Each traffic light object ncis created specifically for each crossing. Here, "crossing" means that vehicle 1 from the plurality of vehicles travels along the roadway section Fb, approaches traffic light A, and then crosses traffic light A. During such a crossing c, vehicle 1 records, in particular continuously, the positions and phase profiles of the individual traffic lights tp 1 -tp 5 using the recording device 3.

[0059] Each of the traffic light objects nc , in particular the traffic light position of the corresponding traffic lights tp 1 -tp 5 represented by it, can be represented as a point (traffic light object point) pn represented in a global coordinate system. The totality of all traffic light objects nc is therefore a point cloud { pn} with n=1....N can be represented in a global coordinate system K.

[0060] For the Figure 2In the method presented, the terms "traffic light objects" and "traffic light object points" can be used interchangeably, since the traffic light object points correspond to a representation of the traffic light objects in the global coordinate system K.

[0061] It is possible to represent the traffic light object points in a "semi-global" coordinate system that is valid for a predetermined area around a given reference point (origin of the "semi-global" coordinate system). This predetermined area can extend several hundred meters from the reference point in all possible directions, especially east, north, and upward (away from the Earth's surface).

[0062] The traffic light object points pn may exhibit an undesirably high scatter in a longitudinal direction along a course of the road section Fb. Therefore, the traffic light object points pnto create the traffic light image. This pre-filtering is described below.

[0063] In a step S102, a course d suit of the road section Fb. This is the road leading to traffic light A. For this purpose, movement history data r traj of the vehicle 1, which represent at least one of the position, the position history and the orientation of the vehicle 1. In other words, the history d suit of the road section Fb is from the movement history data r traj The course can be determined d suit of the road section Fb may also be representative of an orientation of vehicle 1.

[0064] The course r traj of the road section Fb, as the traffic light objects nc also, represented in the global coordinate system K.

[0065] It is possible that step S101 occurs before, after or simultaneously with step S102.

[0066] In a step S103, the traffic light object points pn in a longitudinal direction d traj_intersec of the course d suit of the road section Fb, where the longitudinal direction d traj_intersec the course d suit of the road section Fb on the traffic light line. The traffic light line is the position along the road section Fb at which the traffic light A is located. This can, for example, be the position along the course d suit of the road section where (in the direction of travel of vehicle 1) the vehicle first stops at a traffic light object nc or traffic light point pnpasses by. In particular, the traffic light line can be determined specifically for each crossing (and thus also for each vehicle). For the method described here, a traffic light line spanning all crossings can also be formed by calculating the average of the traffic light lines from each crossing. The traffic light line is therefore also determined / retrieved in step S103.

[0067] Since the traffic light object points pn and the course d suit of the road section Fb in the coordinate system K (as vectors), the projection is carried out for each traffic light object pn according to the following formula: x ′ p n = p → n T d → traj _ intersec

[0068] In a step S104, those traffic light object points pn are discarded which are in the longitudinal direction d suit _ intersec more than two standard deviations σ x'pn of the median value over all projected traffic light object points { x′ pn} So the projected traffic light object points { x′ pn} that meet the following conditions: x ′ p n > median n x ′ p n + 2 σ x ′ p n

[0069] In other words, traffic light object points pn taking into account a scatter (in the longitudinal direction) of all traffic light object points { pn} are sorted out. Alternatively, it is also possible to provide a predetermined limit value, which in particular does not depend on the standard deviation σ x'pn This predetermined limit can, for example, be a fixed (absolute) value.

[0070] The discarded traffic light object points pn represent with a high probability traffic lights that are not located on the traffic light line of traffic light system A valid for the road section Fb, such as pedestrian traffic lights, traffic lights for road sections from other directions)

[0071] Steps S102, S103 and S104 represent the above-mentioned pre-filtering of the traffic light object points pn.

[0072] Furthermore, the traffic light objects remaining after pre-filtering pn divided into groups using a clustering method. A clustering method is a method for discovering similarity structures in (large) data sets, such as the traffic light objects presented here. pn , is meant. The groups of "similar" objects found by the clustering process are called clusters.

[0073] In a step S105, the traffic light object points pn projected onto an image plane E which is perpendicular to the course d suit of the road section Fb at the level of the traffic light line. In other words, the image plane E is the plane perpendicular to the direction of travel of vehicle 1 and along the course d suit of the carriageway section Fb at the level of the traffic light line.

[0074] For projection onto the image plane E, each traffic light object point is projected onto a transverse direction d traj _ intersec ⊥ of the course d suit of the road section Fb, where the transverse direction d traj _ intersec ⊥ of the course d suit of the road section Fb on the traffic light line and lies in a plane of the road section Fb. Heights of the traffic lights tp 1 -tp 5 are also represented by the traffic light object points pn and can be adopted (without being projected in a direction). Accordingly, each traffic light object projected onto the image plane E p' n as follows by a 2D vector of the coordinate system K: p → ′ n = p → n T d → traj _ intersec p → n T k →

[0075] Here is k the canonical basis vector of the coordinate system K.

[0076] In a step S106, a (first) clustering method is applied to the traffic light object points projected in the direction of travel p' n This allows the images projected in the direction of travel Ampeloobjectpoint p' n be grouped to detect / determine traffic light groups. The traffic light object points p' nare clustered according to their position (data).

[0077] Different clustering methods can be used for this purpose. Typically, the well-known algorithm “Density-Based Spatial Clustering of Applications with Noise” (DBSCAN) is used. This density-based algorithm can detect multiple clusters. Noise points are ignored and returned separately.

[0078] The algorithm forms clusters of so-called "densely connected" points, i.e., points that are located no further than a specified distance (neighborhood length) from a "core point" in the same cluster t. A core point is a point that is closer than the neighborhood length ε to at least a predetermined minimum number minPts of other core points in the same cluster t.

[0079] For the present method, a neighborhood length ε 1 is two meters and the predetermined minimum number minPts = 3. In this case, a comparatively high neighborhood length ε 1 =2m is chosen in order to take into account the comparatively high scatter in the traffic light objects p' n , especially with regard to their positions (in the coordinate system).

[0080] Other values are also possible to adapt the clustering procedure to different circumstances.

[0081] The result of the clustering procedure from step S106 are clusters t with t=1...T from the traffic light object points projected in the direction of travel p' n . Each of these clusters t represents a traffic light object group that is representative of a "physical" traffic light tp 1 -tp 5.

[0082] In a step S107, the traffic light objects belonging to a traffic light cluster t, i.e. a traffic light t, are referenced with the corresponding index t, so that each projected traffic light object point with p' nc,t The projected traffic light object point p' nc,t can therefore be assigned to corresponding traffic light cluster t and corresponding crossings c.

[0083] It should be noted that a traffic light group t can comprise more than one traffic light object (and thus a traffic light position), which can have its origin(s) in the same crossing c. The reason for this is that double detection by the detection device 3 is possible, whereby these traffic light objects may be at least partially temporally overlapping or non-temporally overlapping.

[0084] In a step S108, the results of the first clustering procedure are filtered.

[0085] For this purpose, traffic light object points p' n, which cannot be assigned to any traffic light group t and thus represent noise, are discarded.

[0086] Alternatively or additionally, in step S108, traffic light cluster t with a number of traffic light object points p' n that is smaller than a predetermined minimum number, e.g., one-fifth of the number of the second largest traffic light cluster, is discarded. The traffic light object points p' n of the relevant traffic light cluster t are also discarded. This ensures that no traffic light cluster t is taken into account for the further procedure that consists of scattered traffic light object points that do not correspond to any physical traffic light t. p' n are educated.

[0087] Alternatively or additionally, in step S108 a principal component analysis is carried out for each traffic light cluster t.

[0088] Due to the comparatively high first neighborhood length ε 1 =2m, two closely adjacent traffic lights t may be combined into a single cluster t by the clustering method. As a result, the two closely adjacent traffic lights t are not recognized as such, but rather are represented by a traffic light cluster t and thus only as one traffic light t. Such a traffic light cluster t can be determined using principal component analysis.

[0089] Principal component analysis is a multivariate statistical technique used to structure, simplify, and visualize large data sets by approximating a large number of statistical variables using a smaller number of meaningful linear combinations (principal components).

[0090] For the principal component analysis of a traffic light cluster t, all traffic light object points belonging to the corresponding traffic light cluster t p' ntThe following formula can be used for this: p ′ n t = ∪ c = 1 C p ′ n c , t

[0091] In the formula above, c represents the corresponding crossing from c=1... C.

[0092] From this traffic light object point cloud { p' nt} the principal components (principal directions) and the corresponding (maximum) eigenvalues are determined using principal component analysis λ maxt An eigenvalue λ maxt corresponds to the length of a corresponding principal component. Principal component analysis is used to determine how far apart a traffic light cluster t is.

[0093] This allows those traffic light clusters to be identified as "double clusters" whose principal component is longer than a predetermined value. In other words, double clusters exist when eigenvalues λ maxt of the double cluster exceed a predetermined eigenvalue threshold.

[0094] In this case, the predetermined eigenvalue limit can be twice the median value of all eigenvalues λ maxt across all traffic light clusters t. Therefore, the following condition can be applied to identify the double clusters: λ max t > 2 median t λ max t

[0095] It is also possible to use other predetermined eigenvalue limits. For example, the predetermined eigenvalue limit can be a different factor of the median value of the eigenvalues λ maxt across all traffic light clusters than the above double factor.

[0096] The clustering procedure (second clustering procedure) is applied again to the identified double clusters. Specifically, the DBSCAN algorithm is applied again, but with a neighborhood length ε 2 that corresponds to half the first neighborhood length ε 1 of the first clustering procedure. In this case, the neighborhood length ε 2 = 1m. The neighborhood length ε 2 of the second clustering procedure can also have a different value, as long as it is shorter than the first neighborhood length ε 1 . The important thing here is that the second clustering procedure again detects "subclusters" within the double clusters.

[0097] The resulting "sub-clusters" are treated as traffic light clusters t. Traffic light object points resulting from the DBSCAN algorithm with the neighborhood length ε 2 of the second clustering method { p' n} that cannot be assigned to a "sub-cluster" are discarded.

[0098] In step S109, all crossing-specific data records that contain fewer than a predetermined minimum number of traffic light objects after the above steps are discarded. Thus, a check is carried out to determine whether, after the previous steps, only a predetermined minimum number of traffic light objects remains for a crossing c.

[0099] Here, this predetermined minimum number is two. In other embodiments, a different predetermined minimum number of traffic light objects is also possible. It is important that this minimum number is chosen in such a way that a reliable assignment of the traffic light object points is ensured. p' nt to the traffic light clusters t.

[0100] If a traffic light cluster t no longer has any traffic light objects as of step S109, these "empty" traffic light clusters are also discarded.

[0101] In step S110, the traffic light image 7 is created. For this purpose, a spatial average is calculated from each of the remaining traffic light clusters t, which then represents a position of the physical traffic light t represented by the corresponding traffic light cluster t. The traffic light object points remaining from the previous steps S101 to 108 p' nt can be assigned to the physical traffic lights tp 1 -tp 5.

[0102] An example of the result of the previously described procedure is shown in Figure 3 Here, traffic light image 7 is shown that traffic light system A consists of Figure 1 represents the traffic light objects nc,t from all crossings c are represented as a (cross-shaped) point cloud { p' nt} It can be seen that in the traffic light image 7 the traffic light objects nc,tare grouped into four clusters t=1 to t=4 (clusters t 1 -t 4 ). Each cluster corresponds to one of the traffic lights tp 1 to tp 4 . The hatched diamonds represent the centers of the clusters t (spatial means), each of which represents a spatial traffic light position of the corresponding physical traffic light.

[0103] The traffic light objects recorded for traffic light tp 5 are assigned to the cluster for traffic light tp 4. This is because traffic light tp 5 (as a right-turn light) is located very close to or even adjacent to traffic light tp 4. The Figure 3 The horizontal axis (x-axis) represents a direction transverse to the road section and the vertical axis (y-axis) represents a direction perpendicular to the earth's surface towards the sky. Figure 3 corresponds, so to speak, to a view from vehicle 1 onto traffic light system A. The axis values are given in meters. For example, traffic light cluster t 2 is located approximately 4 m to the left and 7 m above vehicle 1.

[0104] In the Figure 4 For further explanation, another traffic light image 7' of the traffic light system A is shown as an example and schematically. In the traffic light image 7', a variety of traffic light objects nc,t which are representative of the traffic lights tp 1 -tp 5 of traffic light system A.

[0105] As described above, the traffic light objects nc,t recorded specifically for each crossing. As a rule, "crossing" means that vehicle 1 from the plurality of vehicles travels along the roadway section Fb, approaches the traffic light system A and then crosses it. During a crossing c, vehicle 1 records, in particular continuously, the positions and phase profiles of the individual traffic lights tp 1 -tp 5 . By means of the Figure 2 The procedure described is used to calculate the totality of the traffic light objects included in traffic light image 7' nc,tTraffic light clusters t 1 , t 2 , t 3 , t 4 are formed, whereby the traffic light clusters t 1 -t 4 correspond to the physical traffic lights tp 1 -tp 5 of the traffic light system A.

[0106] From the traffic light image 7' it can be determined that in this example the traffic light objects nc,t from four crossings c=1...4. It should be noted that these four crossings c=1...4 may not only have been carried out by the same vehicle 1, but also by different vehicles from the plurality of vehicles. In the Figure 4 are the traffic light objects nc,t in different forms, where each form represents a corresponding crossing c. It can be seen that a crossing c=1 for the traffic light tp 1 and the traffic light tp 3 results in a traffic light object n 1.1 or n 1,3 and assigned to the traffic light clusters t 1 , t 3 . For the same crossing c=1, two traffic light objects are recorded for the traffic light tp 2 and the traffic light tp 4 n1.2 or n 1,4 and assigned to the traffic light cluster t 2 , t 4 .

[0107] The fact that more than one traffic light object is generated for a traffic light during a single pass can be due, among other things, to the detection device 3 of the corresponding vehicle incorrectly detecting the corresponding traffic light and creating and assigning multiple traffic light objects for this traffic light. It is also possible that during detection by the detection device 3, a line of sight between the detection device 3 and the corresponding traffic light is interrupted by an obstacle, so that after the obstacle is removed, the corresponding traffic light is detected again by the detection device 3, and therefore a new traffic light object is generated and assigned to this traffic light.

[0108] It should be noted that the Figure 4 shown different shapes of the traffic light objects nc,tare merely for clarification and the disclosure is not limited to the fact that the traffic light objects nc,t are represented in a special form in the traffic light image 7'.

[0109] In Figure 5 A method for grouping the traffic lights tp 1 -tp 5 is shown. The groups are formed in such a way that each traffic light within a (traffic light) group has the same phase behavior, or a similar phase progression.

[0110] In a step S201, the traffic light image 7 or the traffic light objects nc,t which represents the traffic light system A. As described above, the traffic light image 7 shows the traffic light objects recorded over several crossings c nc,t for each traffic light tp 1 -tp 5 of the traffic light system A. The Figure 5 The method shown does not specifically require traffic light image 7 of traffic light system A (as the input data set). Other traffic light images representing traffic light systems can also be used.

[0111] In a step S202, inconsistent traffic lights are identified. It should be noted that the physical traffic lights tp 1 -tp 5 are represented by corresponding traffic light clusters t 1 -t 4 in the traffic light image 7.

[0112] A traffic light is inconsistent if the phase progression of individual traffic light objects assigned to the same traffic light appears inconsistent with each other. This usually indicates that there is a (right or left) turn signal at the affected traffic light that only occasionally illuminates. These turn signals are usually located so close to the affected traffic light that the Figure 2 The procedure shown incorrectly assigns the traffic light object representing the turning light to the traffic light in question. This situation occurs, for example, in the Figure 1 shown traffic light system A. Here, the traffic light tp 5 is located close to the traffic light tp 4, so that the Figure 2The method shown does not detect two traffic light clusters for the traffic lights tp 4 and tp 5, but only the traffic light cluster t 4 . In other words, the two traffic lights tp 4 and tp 5 are Figure 2 The procedures shown are not differentiated.

[0113] In the exemplary traffic light image 7', the traffic light object n 2,4 and traffic light object n 2.5 of the traffic lights tp 4 and tp 5 are so close to each other that they are assigned to the same traffic light cluster t 4. Therefore, the Figure 4 displayed traffic light object n 2.5 (the traffic light object created for a crossing c=2 for the traffic light t 5 ) in the traffic light cluster t 4 .

[0114] In order to counteract this circumstance of inconsistent traffic lights or traffic light clusters, for each traffic light cluster t 1 to t 4 for each crossing c all possible traffic light object pair combinations are determined and for each of the determined traffic light object pair combinations a correlation factor ρ nc,t,n'c,tThe correlation factors ρ nc,t,n'c,t ( ρ obj ) are formed as follows: ρ n c , t , n ′ c , t = Tm n c , t , n ′ c , t Tt n c , t , n ′ c , t

[0115] Here, nc,t a first traffic light object of the traffic light object pair and n' c,t a second traffic light object of the traffic light object pair. Tm nc,t,n'c,t and Tt nc,t,n'c,t represent the overlap duration or match duration between the first traffic light object nc,t and the second traffic light object n' c,t of the traffic light object pair.

[0116] All traffic light objects assigned to a crossing c are calculated taking into account the corresponding correlation factors ρ nc,t,n'c,tclustered. In this case, this is usually done using a hierarchical clustering algorithm. The resulting clusters are counted. Then, for each traffic light cluster t 1 to t 4, the average of the number of resulting clusters across all crossings c is determined. If this determined average exceeds a predetermined threshold, the traffic light cluster is considered inconsistent. The predetermined threshold can be, for example, 1.2.

[0117] In the present example of traffic light image 7, the traffic light cluster t 4 represents an inconsistent traffic light cluster.

[0118] In a step S203, the phase curves of all traffic light objects nc , especially those of the traffic light clusters rated as consistent, are checked for plausibility. In this case, those traffic light objects ncfor the following steps of the procedure, whose phase progressions show the following sequences of traffic light phases within their phase progression: yellow phase to red phase, red-yellow phase to green phase, green phase or red-yellow phase to a yellow phase, red phase or a yellow phase to a red-yellow phase. Implausible traffic light objects nc are discarded

[0119] In a step S204, all traffic light pair combinations are formed from the consistent traffic light clusters t 1 , t 2 , t 3 . In this case, the traffic light pair combinations t 1 / t 2 , t 1 / t 3 , t 2 / t 3 are formed.

[0120] Subsequently, in step S205, an overlap duration Tt and a match duration Tm between all traffic light objects are determined across all crossings. nc,t , which are assigned to the first traffic light cluster of the traffic light pair, and all traffic light objects nc,t ,which are assigned to the second traffic light cluster of the traffic light pair, are summed up and a correlation factor ρ AP between the first and second traffic lights of the traffic light pair. With reference to Figure 7 The terms "overlap duration" and "match duration" will be described in more detail later.

[0121] Step S205 is described as an example for the traffic light pair t 1 / t 2. For the calculation of the overlap duration across crossings Tt t 1, t 2 (Tt AP ) for the traffic light pair consisting of traffic light cluster t 1 and traffic light cluster t 2 the following formula applies: Tt t 1 , t 2 = ∑ c = 1 C ∑ n c , t 1 = 1 N c , t 1 ∑ n c , t 2 = 1 N c , t 2 Tt n c , t 1 , n c , t 2

[0122] Here c stands for the respective crossing from c=1...C and n c,t 1 and n c,t 2 for the respective traffic light object of the traffic light cluster t 1 or traffic light cluster t 2 , where N c,t 1 and N c,t 2 the number of traffic light objects nc,twhich were created / recorded from a crossing c for the corresponding traffic light tp 1 or traffic light tp 2. Tt n c , t 1 , n c , t 2 Tt n stands for the crossing-specific overlap duration between a traffic light object belonging to the traffic light cluster t 1 n c,t 1 and a traffic light object belonging to the traffic light cluster t 2 n c,t 2 .

[0123] For example, from the exemplary traffic light image 7' Figure 4 determine that for the crossing c=1 for the first traffic light tp 1 only one (circular) traffic light object n 1,1 and for the second traffic light tp 2 two (circular) traffic light objects n 1,2 . For the second crossing c=2, two (triangular) traffic light objects are located for the first traffic light tp 1 n 2,1, while for the second traffic light tp 2 only one (triangular) traffic light object n 2,2 is present. For the crossing c=3, there is a (star-shaped) traffic light object for both traffic lights t 1 ,t 2 n 3.1 , n3.2 before.

[0124] To calculate the match duration Tm t 1, t 2 (Tm AP ) for the traffic light pair consisting of traffic light cluster t 1 and traffic light cluster t 2 the following formula applies: Tm t 1 , t 2 = ∑ c = 1 C ∑ n c , t 1 = 1 N c , t 1 ∑ n c , t 2 = 1 N c , t 2 Tm n c , t 1 , n c , t 2 Tm n c , t 1 , n c , t 2 Tm n stands for the crossing-specific match duration between a traffic light object belonging to the traffic light cluster t 1 n c,t 1 and a traffic light object belonging to the traffic light cluster t 2 n c,t 2 .

[0125] A correlation factor can then be ρ t 1 ,t 2 (general ρ AP ) for the traffic light pair t 1 / t 2 according to the following: ρ t 1 , t 2 = Tm t 1 , t 2 Tt t 1 , t 2

[0126] In step S206, all consistent traffic light clusters are grouped based on the correlation factors of all traffic light pair combinations. The traffic lights are grouped such that all traffic lights within a group exhibit the same phase progression. Grouping is performed using appropriate clustering methods, such as the hierarchical clustering algorithm. The resulting groups, called traffic light groups, are referenced with an index g=1...G. Each consistent traffic light cluster is assigned to one of these traffic light groups.

[0127] In step S207, the traffic light clusters t* identified as inconsistent are assigned to a traffic light group g. For each crossing c, an overlap duration Tt nc,t*,g and a match duration Tm nc,t*,g for a traffic light object nc,t* in an inconsistent traffic light cluster t* with respect to all traffic light objects nc,tg in a traffic light group g.

[0128] The crossing-specific overlap duration Tt nc,t*,g between the traffic light object nc,t* and all traffic light objects nc,tg from traffic light group g as follows: Tt n c , t ∗ , g = ∑ t g = 1 T g ∑ n c , t g = 1 N c , t g Tt n c , t ∗ n c , t g

[0129] Here, T g represents the number of traffic light clusters in the respective traffic light group g. Furthermore, nc,tg for the respective traffic light object assigned to the traffic light cluster tg, where N c,tg represents the number of traffic light objects within a crossing c for the corresponding traffic light cluster tg. Tt n c , t ∗ n c , t g stands for the crossing-specific overlap duration between a traffic light object belonging to the inconsistent traffic light t* nc,t* and a traffic light object belonging to the traffic light cluster tg nc,tg .

[0130] The crossing-specific overlap duration Tm nc,t*,g per crossing c is determined as follows: Tm n c , t ∗ , g = ∑ t g = 1 T g ∑ n c , t g = 1 N c , t g T m n c , t ∗ , n c , t g Tm n c , t ∗ , n c , t g stands for the crossing-specific match duration between a traffic light object belonging to the inconsistent traffic light cluster t* nc,t* and all traffic light objects belonging to traffic light group g nc,tg .

[0131] Then, for each inconsistent traffic light t* and for each formed traffic light group g, a set of cross-crossing Γ g,t* with all traffic light objects nc,t* the inconsistent traffic light t* is formed according to the following formula: Γ g , t * = ∪ c = 1 C n c , t * Tt n c , t ∗ , g > 0 ∧ Tm n c , t ∗ , g Tt n c , t ∗ , g > 0.99

[0132] One can therefore see that the sentence Γ g,t* associated traffic light objects of the inconsistent traffic light t* to the traffic light objects nc,g the traffic light group g a minimum correlation factor Tm n c , t ∗ , g Tt n c , t ∗ , g of 0.99. The value of the minimum correlation factor is exemplary and can be adjusted.

[0133] For the traffic light objects from the set Γ g,t* An overlap duration is set across all crossings and traffic light objects Tt t*,gand a match duration Tm t*,g determined according to the following formulas. Tt t * , g = ∑ c = 1 C ∑ n c , t * = 1 N c , t * Tt n c , t * , g Tm t * , g = ∑ c = 1 C ∑ n c , t * = 1 N c , t * Tm n c , t * , g mit n c , t * ∈ Γ g , t *

[0134] In the above two formulas, the following also applies: n c,t* ∈ Γ g,t* . So only traffic light objects n c,t* the inconsistent traffic light t* which have or exceed the above-mentioned minimum correlation factor to the traffic light group g.

[0135] For each inconsistent traffic light t*, a corresponding traffic light group g t * * determined so that the number of traffic light objects Γ g,t* the inconsistent traffic light t* is highest, where the traffic light objects n c,t* ∈ Γ g,t* with the phase of the corresponding group g t * * a high correlation factor ( Tm n c , t * , g Tt n c , t * , g > 0 , 99 ).

[0136] The following formula is used: g t * * = argmax g Γ g , t *

[0137] The inconsistent traffic light t* is assigned to this traffic light group g t * * assigned if the number of correlated traffic light objects Γ g t * * , t * from the traffic light objects n c,t* the inconsistent traffic light t* and traffic light objects n c , g t * * the corresponding traffic light group g t * * equal to or greater than a predetermined fraction of all (correlated and non-correlated) traffic light objects n c,t* of the inconsistent traffic light t*. The predetermined fraction can be, for example, one-third.

[0138] Only the correlated traffic light objects n c,t* ∈ Γ g,t* the inconsistent traffic light t* of the traffic light group g t * * The remaining (non-correlated) traffic light objects n c,t* ∉ Γ g,t* correspond with high probability to (phase) sightings of the turning traffic light t 5 , which cannot be further interpreted.

[0139] In Figur 6 is a method for determining a traffic light phase of a traffic light or a traffic light group of a traffic light system. The method is then performed for a traffic light from the traffic lights tp 1 - tp 5 or a traffic light group from the traffic light groups g.

[0140] For example, the information represented by traffic light objects can be used for the process. Alternatively, the information required for the process can also be provided / represented in a different form than traffic light objects.

[0141] In particular, the method is performed when vehicle 1 is just before or at the stop line of traffic light system A (corresponding to the stop line at an intersection). The stop line may be known, for example, from map information. Alternatively, the stop line can be determined as a median value across all positions along the course of the roadway section (or along the trajectory of vehicle 1) at which vehicle 1 passes the traffic light objects.

[0142] In any case, vehicle 1 is in a position where the vehicle-mounted detection device 3 can no longer detect the traffic light system A and the traffic lights. This is primarily due to a limited aperture angle of a camera device of the detection device 3. However, the time when crossing or stopping at the stop line is particularly relevant for traffic light lane assignment.

[0143] In a step S301, a time T event is retrieved at which the vehicle 1 either crosses the stop line or is at it. This can be derived from the vehicle-side detection system 3 and, in particular, in combination with information from the traffic light objects n c,t .

[0144] If vehicle 1 passes this stop line at a speed greater than a predetermined speed, e.g., 3 km / h, this is interpreted as crossing the stop line. If vehicle 1 falls below a speed of 1 km / h for the first time at the stop line, this is interpreted as stopping at the stop line. "At the stop line" refers to a predetermined distance of vehicle 1 (in the direction of travel) from the stop line of up to at least 25 m, preferably 15 m. This can refer to the distance from the front end of vehicle 1 to the stop line. Other predetermined distances are also conceivable, provided they are suitable for indicating that the vehicle is stopping at the stop line.

[0145] In a step S302, a first detection time T i is retrieved at which the vehicle 1 detects a traffic light for the first time.

[0146] A time interval ΔT i between the time T i of the first sighting of the traffic light and the event time T event can be determined.

[0147] In a step S303, a last detection time T e at which the vehicle last detects the traffic light is retrieved.

[0148] A time interval ΔT e between the time T e of the last sighting of the traffic light and the event time T event can be determined. The time interval ΔT e is set to 0 if the traffic light is still visible at the event time T event.

[0149] In a step S304, a phase profile of the traffic light is retrieved, which extends from the first detection time T i to the last detection time T e . From the retrieved phase profile, a traffic light phase C g of the traffic light can be determined, which is present at time T i . As a rule, the traffic light phase C g can have at least the following values: green, yellow, red, and red and yellow. Thus, the following traffic light phases C g can occur: red phase, yellow phase, red phase, green phase, and red-yellow phase.

[0150] If a traffic light phase Cg of a traffic light group is retrieved, it may happen that traffic lights in this traffic light group accept / return conflicting traffic light phases instead of a common traffic light phase. In this case, the traffic light phase Cg for this traffic light group is set to "invalid."

[0151] In step S305, a phase change time T switch is determined or retrieved at which a phase change of the traffic light occurs or was detected. In other words, at time T switch , the traffic light phase C g of the traffic light changes. The phase change time T switch is before the last detection time T e .

[0152] A time interval ΔT switch between the time T switch of the phase change of the traffic light and the event time T event can be determined.

[0153] If there is no phase change or if there are contradictions between two or more traffic lights in a traffic light group, the time T switch and thus also the time interval Δt switch are set to "invalid." If there is no phase change in the retrieved phase curve, the phase change time can be determined (as described later).

[0154] In step S306, a traffic light phase C Δ-< before the traffic light phase change time T switch and a traffic light phase C Δ+< after the traffic light phase change time T switch are determined. The traffic light phases C Δ-< and C Δ+< thus indicate the traffic light phases that occur before and after the time T switch, respectively. The traffic light phases C Δ-< and C Δ+< can assume the traffic light phases described in step S304.

[0155] In step S307, the traffic light phase C g is determined at the event time T event . Specifically, a probability P(C g = red) is determined that the traffic light has a red phase at time T event . In other words, the probability that the traffic light is red (or indicates a red phase) when vehicle 1 stops at the traffic light or passes it is determined. The following applies to the probability P(C g = red): P C g = rot = P C g = rot Δ T e = Δ t e , Δ T i = Δ t i , C g = c , Δ T switch = Δ t switch , C Δ + = c Δ + , C Δ − = c Δ −

[0156] The probability P(C g =rot) is therefore a conditional probability that takes into account the quantities ΔT e , ΔT i , C g , ΔT switch , C Δ-< and C Δ+<.

[0157] The probability P(C g =green) that the traffic light has a traffic light phase C g of "green" at time T event is: P C g = gr ü n = = P C g = gr ü n Δ T e = Δ t e , Δ T i = Δ t i , C g = c , Δ T switch = Δ t switch , C Δ + = c Δ + , C Δ − = c Δ − = = 1 − P C g = rot

[0158] The probability P(C g =green) is also a conditional probability that takes into account the quantities ΔT i , ΔT e , C g , ΔT switch , C Δ-< and C Δ+<.

[0159] To calculate the probabilities P(C g = red) or P(C g = green), three cases I, II, and III can occur. These cases depend on the time intervals ΔT e and / or ΔT switch.

[0160] In case I, ΔT e =0s. The detection device 3 still saw the traffic light at time T event . Accordingly, the traffic light phase C g is known, so the probability P(C g =red) is: P C g = rot Δ T e = 0 , C g = c = 1 , 0 , 0 , 8 , 0 , 7 , c = rot c = gr ü n c = gelb c = gelb + rot

[0161] The probability P(C g = red) of 0.8 for the traffic light phase c = yellow allows a yellow traffic light to be interpreted as green or red, whereby this value of 0.8 is driver-dependent. The same applies to the probability P(C g = red) of 0.7 for c = yellow + red. The probabilities for c = yellow and c = yellow + red are examples here. Alternatively, these values can be determined on a driver-specific basis. In other words, the probabilities for c = yellow and c = yellow + red can be determined / ascertained depending on a driver's driving behavior.

[0162] In case II, the time interval ΔT e is greater than 0 s and the time interval ΔT switch has a valid value (and thus a value greater than 0 s). Thus, the traffic light phase C g at time T event can be derived from the last detected phase change at time T switch.

[0163] For this purpose, the phase durations of the traffic light phases, which are usually unknown, are taken into account by modeling them as random variables. Thus, a reference traffic light cycle is modeled. In the model, T cycle represents the cycle time of the reference traffic light cycle. A cycle time corresponds to the time it takes a traffic light to complete a complete switching process. For example, a complete switching process can be a traffic light phase sequence with the traffic light phases in the order "green," "yellow," "red," and "yellow and red."

[0164] The orbital time T cycle is: T cycle ∼ U t cycle _ min t cycle _ max

[0165] Here, t cycle_min is a minimum cycle time and t cycle_max is the maximum cycle time.

[0166] For the reference traffic light cycle, it is assumed that the cycle time is evenly distributed in the above value range [t cycle_min ; t cycle_max ].

[0167] For example, t cycle_min could be 30s and t cycle_max 120s. These values are taken from the Traffic Signal Guidelines, which are a set of rules applicable in Germany. Different values may be specified for other countries.

[0168] For a phase duration X red in which the reference traffic light cycle shows red, the following applies: X rot ∼ U rot min rot max

[0169] Here, too, it is assumed for the reference traffic light cycle that the phase duration X rot is uniformly distributed within the value range [rot min; rot max]. Typically, rot min and rot max indicate the minimum and maximum fractions of the cycle time T cycle at which the reference traffic light cycle indicates "red." For example, rot min can be 0.3 and rot max can be 0.7. Alternatively, the value range for the phase duration X rot can also be specified by absolute minimum and maximum times in seconds.

[0170] For case II, the probability P(C g = red) is: P C g = rot Δ T switch = Δ t switch , C Δ + = c Δ + , T cycle = t cycle , X rot = x rot = = min Δ t switch ′ t trans 1 , c Δ + = rot ∨ gelb , Δ t switch ′ < x rot ∗ t cycle 0 , c Δ + = rot ∨ gelb , Δ t switch ′ ≥ x rot ∗ t cycle 0 , c Δ + = gr ü n ∨ rot + gelb , Δ t switch ′ < 1 − x rot ∗ t cycle min Δ t switch ′ − 1 − x rot ∗ t cycle t trans 1 , c Δ + = gr ü n ∨ rot + gelb , Δ t switch ′ ≥ 1 − x rot ∗ t cycle

[0171] Here, Δ t switch ′ the cycle-corrected time interval that has elapsed since the last phase change. For Δ t switch ′ applies: Δ t switch ′ = Δ t switch mod t cycle

[0172] In particular, the cycle-corrected time interval Δ t switch ′ used to catch the case that more than one complete cycle of the traffic light has elapsed after the time T e . It is also possible to use only the (uncorrected) time interval Δt switch to be used for case II.

[0173] Furthermore, t trans is a predetermined, particularly fixed, parameter corresponding to a transition time after the traffic light changes to "red." During this transition time t trans , a driver of vehicle 1 behaves as if the traffic light were still "green." During this transition time t trans , there is a probability other than 0 that the current phase of the traffic light will be "interpreted" as green. The transition time t trans typically has a value of a few seconds. For example, the transition time t trans can assume a value between 0 s and 5 s, and in particular 3 s. Furthermore, an additional transition time after a phase change to "yellow" can be taken into account. The additional transition time is typically longer than the transition time t trans , and the probability that a driver will cross the traffic light during the additional transition time is higher.

[0174] The case distinctions from Case II are explained below.

[0175] If the traffic light phase C Δ+< present after the phase change is a red or yellow traffic light phase, the time interval ΔT switch (or the cycle-corrected time interval Δ t switch ′ ) is compared with the red phase duration X red of the reference traffic light cycle. Here, the red phase duration X red of the reference traffic light cycle is given as a fraction X red of the cycle time t cycle.

[0176] If the time interval ΔT switch is smaller (shorter) than the red phase duration X red of the reference traffic light cycle, it is assumed that the red traffic light phase is present with a probability P(C g = red), where the probability P(C g = red) is the smaller value of Δ t switch ′ t trans 1 It is therefore assumed that the traffic light has not switched to the next traffic light phase after the last detection time T e , so that a red traffic light phase exists at the event time T event .

[0177] However, if the time interval ΔT switch is greater (longer) than or equal to the red light phase X red of the reference traffic light cycle, it is assumed that the traffic light has switched to the next (i.e., green) phase between the last detection time T e and the event time T event , so that a green light phase exists at the event time T event . The probability P(C g = red) is then 0.

[0178] The same applies to the further case distinctions in Case II.

[0179] Note that the reference traffic light cycle does not need to consider the transition time t trans . Thus, the transition time t trans can be ignored, resulting in a probability of 1 for the first and fourth case distinctions from Case II. The following then applies: P C g = rot Δ T switch = Δ t switch , C Δ + = c Δ + , T cycle = t cycle , X rot = x rot = = 1 , c Δ + = rot ∨ gelb , Δ t switch ′ < x rot ∗ t cycle 0 , c Δ + = rot ∨ gelb , Δ t switch ′ ≥ x rot ∗ t cycle 0 , c Δ + = gr ü n ∨ rot + gelb , Δ t switch ′ < 1 − x rot ∗ t cycle 1 , c Δ + = gr ü n ∨ rot + gelb , Δ t switch ′ ≥ 1 − x rot ∗ t cycle

[0180] The probability P(C g =rot) given for Case II can be calculated for all possible orbital periods T cycle and all possible phase durations X rot. Thus, a total probability P(C g =rot) for all orbital periods T cycle and phase durations X rot can be determined as follows: P C g = rot Δ T e = Δt e , Δ T switch = Δ t switch , C Δ + = c Δ + = = ∫ t cycle _ min t cycle _ max ∫ max rot min Δ t switch ′ − Δ t e t cycle rot max P C g = rot Δ T switch = Δ t switch , C Δ + = c Δ + , T cycle = t cycle , X rot = x rot ∗ P X rot = x rot ∗ P T cycle = t cycle dx rot dt cycle

[0181] The term P ( C g = rot | ΔT switch = Δ t switch , C Δ+< = c Δ + < , T cycle = t cycle ,X rot = x rot ) can be determined as indicated above.

[0182] P ( X rot = x rot ) is the probability that the phase duration X rot takes a certain value X rot from the above-mentioned value range [rot min ; rot max ] for the phase duration. P ( T cycle = t cycle ) is the probability that the orbital period T cycle assumes a specific value t cycle from the above-mentioned range of values [t cycle_min ; t cycle_max ] for the orbital period. In the case of a uniform distribution of the orbital periods and phase durations modeled as random variables, as is the case here, the probability is the same for each value from the range.

[0183] To calculate the overall probability, the integrals can be approximated, for example, by summing. Other approximation methods are also possible.

[0184] In Case III, neither of the two previous Cases I and II applies. This means that in Case III, only the last seen / detected traffic light phase is known. Accordingly, it is unknown how long this traffic light phase has been present.

[0185] Therefore, the probability P(C g = red) is determined by taking into account the relative duration of the green and red phases and the elapsed time between time T i and time T e as a function of the last observed phase. This results in four subcases: III.a, III.b, III.c, and III.d. The above explanations for Case II apply accordingly.

[0186] For case III.a, where the last seen phase C g = red, the following applies to the probability P(C g = red): P C g = rot = P C g = rot Δ T e = Δ t e , Δ T i = Δ t i , C g = rot , T cycle = t cycle , X rot = x rot = ∫ Δ t i Δ t e + t cycle P C g = rot Δ T switch = Δ t switch , C g = rot , T cycle = t cycle , X rot = x rot ∗ P Δ T switch = Δ t switch Δ T e = Δ t e , Δ T i = Δ t i , C g = rot , T cycle = t cycle , X rot = x rot d Δ t switch

[0187] Furthermore: P C g = rot Δ T switch = Δ t switch , C g = rot , T cycle = t cycle , X rot = x rot = 1 , Δ t switch mod t cycle < x rot ∗ t cycle 0 , Δ t cycle mod t cycle ≥ x rot ∗ t cycle

[0188] The phase change time T switch and the corresponding time interval ΔT switch are unknown and are therefore modeled. For this purpose, it is assumed that the phase change time T switch had to occur between one orbital time T cycle before the first acquisition time T i and the last acquisition time T e . Therefore, the probability that the phase change time T switch occurs at a specific time t switch or that the time interval ΔT switch has a specific value Δt switch is as follows: P Δ T switch = Δ t switch Δ T e = Δ t e , Δ T i = Δ t i , C g = rot , T cycle = t cycle , X rot = x rot = 1 Δ t e + Δ t cycle − Δ t i

[0189] In case III.b, the last observed phase C g = green. This corresponds to case III.a, except that the green phase replaces the red phase in case III.a. Thus, the following applies: P C g = rot = P C g = rot Δ T e = Δ t e , Δ T i = Δ t i , C g = gr ü n , T cycle = t cycle , X rot = x rot = ∫ Δ t i Δ t e + t cycle P C g = rot Δ T switch = Δ t switch , C g = gr ü n , T cycle = t cycle , X rot = x rot ∗ P Δ T switch = Δ t switch Δ T e = Δ t e , Δ T i = Δ t i , C g = gr ü n , T cycle = t cycle , X rot = x rot d Δ t switch

[0190] The probability that the red traffic light phase is present at the event time T event if the green traffic light phase was last seen is: P C g = rot Δ T switch = Δ t switch , C g = gr ü n , T cycle = t cycle , X rot = x rot = 0 , Δ t switch mod t cycle < 1 − x rot ∗ t cycle 1 , Δ t switch mod t cycle ≥ 1 − x rot ∗ t cycle

[0191] For the probability that the phase change time T switch occurs at a certain time t switch or that the time interval ΔT switch has a certain value Δt switch, the following applies analogously to case III.a: P Δ T switch = Δ t switch Δ T e = Δ t e , Δ T i = Δ t i , C g = gr ü n , T cycle = t cycle , X rot = x rot = 1 Δ t e + Δ t cycle − Δ t i

[0192] There is the same probability P (Δ T switch = Δ t switch | ...) for cases III.a and III.b, since the cycle time T cycle of the reference traffic light cycle does not change.

[0193] In case III.c, the last seen traffic light phase is yellow. Here, it is assumed that time T e corresponds to time T switch , i.e., T e =T switch . The same applies to the time intervals Δt e =Δt switch . In this case, the probability P(C g =red) is: P C g = rot Δ T e = Δ t e , Δ T i = Δ t i , C g = gelb , T cycle = t cycle , X rot = x rot = 1 , Δ t e ≤ x rot ∗ t cycle 0 , Δ t e > x rot ∗ t cycle

[0194] In case Ill.d, the last observed traffic light phase is "red+yellow." Here, too, it is assumed that time T e corresponds to time T switch , i.e., T e =T switch . Accordingly, the time intervals Δt e =Δt switch also apply. Thus, we get: P C g = rot Δ T e = Δ t e , Δ T i = Δ t i , C g = rot + gelb , T cycle = t cycle , X rot = x rot = 0 , Δ t e ≤ 1 − x rot ∗ t cycle 1 , Δ t e > 1 − x rot ∗ t cycle

[0195] The probabilities P(C g =rot) for cases III.a to III.d can be calculated for all possible orbital periods T cycle and all possible phase durations X rot, similar to case II. Thus, the overall probability is: P C g = rot Δ T e = Δ t e , Δ T i = Δ t i , C g = c = = ∫ t cycle _ min t cycle _ max ∫ max rot min Δ t i − Δ t e t cycle rot max P C g = rot Δ T e = Δ t e , Δ T i = Δ t i , C g = c , T cycle = t cycle , X rot = x rot ∗ P X rot = x rot ∗ P T cycle = t cycle dx rot dt cycle

[0196] Here, too, the integrals can be approximated, for example, by summing, to calculate the overall probability. Other approximation methods are also possible.

[0197] In Figur 7 Two phase curves 9, 9' of two traffic light objects are shown schematically and as examples. The phase curves 9, 9' show the green traffic light phase G, the red traffic light phase R, the yellow traffic light phase Y, and the red-yellow traffic light phase R+Y.

[0198] The combination of the solid lines yields the coincidence duration Tm of the two phase sequences 9, 9', during which they exhibit the same traffic light phase (simultaneously). The combination of the solid lines with the dashed line yields the overlap duration Tt of the two phase sequences 9, 9', during which the phase sequences 9, 9' overlap temporally (regardless of their traffic light phases).

[0199] The overlap duration Tt thus indicates the period of time during which the two phase sequences 9, 9' occur simultaneously, i.e., they overlap in time. The coincidence duration Tm indicates the period of time (within the overlap duration Tt) during which the phase sequences 9, 9' exhibit the same traffic light phase.

[0200] Furthermore, Figur 7 corresponding first acquisition times T i , last acquisition times T e and phase change times T switch of the phase curves 9, 9' are shown. Bezugszeichenliste

[0201] A Traffic light system (A) F1-F3 Lanes Fb Road section G Green traffic light phase tp 1 - tp 5 Traffic light n c Traffic light objects n c,t Traffic light object assigned to a cluster t R red traffic light phase R+Y red-yellow traffic light phase t 1 -t 5 (traffic light / traffic light object) cluster tp 1 -tp 5 physical traffic light Y yellow traffic light phase 1 vehicle 3 detection device 5 radio interface 7 traffic light image 7' traffic light image 9 phase progression 9' phase progression S... process steps

Claims

1. Method for processing data about a traffic light system (A) that controls a road portion and has at least one traffic light (tp1, tp2, tp3, tp4, tps), comprising: - (S101) retrieving traffic light objects (nc) which represent traffic light information about the traffic light system (A), can be represented in a coordinate system (K) by means of a point cloud {p"n} and are recorded by a plurality of vehicles (1) with recording devices (3), wherein the retrieval of the traffic light objects (nc) also includes: - (S102) retrieving a course of the road portion (Fb); - (S103) retrieving a traffic light line which indicates a position of the traffic light system (A) along the road portion (Fb) and determining a longitudinal direction of the road portion (Fb) at the level of the traffic light line; and - (S104) discarding the traffic light objects (nc) which exceed a predetermined limit distance to a median value of all traffic light objects in the determined longitudinal direction; - (S106) identifying at least one cluster in the traffic light objects (nc), wherein an identified cluster (t1, t2, t3, t4) represents a corresponding one of the at least one traffic lights (tp1, tp2, tp3, tp4, tps) of the traffic light system (A); - (S107) assigning the traffic light objects (nc) to the at least one cluster (t1, t2, t3, t4); and - (S110) deriving a spatial position of the at least one traffic light (tp1, tp2, tp3, tp4, tps) on the basis of the assigned traffic light objects (nc,t).

2. Method according to claim 1, wherein the predetermined limit distance depends on a standard deviation of the traffic light objects (nc).

3. Method according to claim 1, wherein the identification of the at least one cluster (t1, t2, t3, t4) also includes: - (S105) projecting the point cloud {p"n} onto a plane that perpendicularly intersects the road portion (Fb) at the traffic light line; and - (S106) identifying the at least one cluster (t1, t2, t3, t4) of traffic light objects (nc) by means of the projected point cloud {p‴n}.

4. Method according to any of the preceding claims, wherein the identification of the at least one cluster (t1, t2, t3, t4) also includes: - (S108) discarding each cluster of which the number of traffic light objects (nc) is less than a predetermined minimum number.

5. Method according to any of the preceding claims, wherein the identification of the at least one cluster (t1, t2, t3, t4) also includes: - (S108) performing a principal component analysis on the at least one cluster (t1, t2, t3, t4); and if an eigenvalue of the at least one cluster (t1, t2, t3, t4) exceeds a predetermined limit eigenvalue: - (S108) identifying at least one sub-cluster in the traffic light objects of the at least one cluster (t1, t2, t3, t4), wherein each identified sub-cluster represents a corresponding one of the at least one traffic lights (tp1, tp2, tp3, tp4, tps) of the traffic light system (A).

6. Method according to any of the preceding claims, wherein the derivation of the spatial position of the at least one traffic light (tp1, tp2, tp3, tp4, tps) includes: - (S110) forming a spatial average over the traffic light objects (nc) which are assigned to the cluster (tp1, tp2, tp3, tp4) that represents the at least one traffic light (tp1, tp2, tp3, tp4, tps).

7. Method according to any of the preceding claims, wherein the identification of the at least one cluster is carried out by means of a first clustering method and / or the identification of the at least one sub-cluster is carried out by means of a second clustering method.

8. Method according to claim 7, wherein at least one of the first and second clustering methods includes the DBSCAN algorithm.

9. Method according to claim 8, wherein a neighborhood length of the first clustering method is greater than a neighborhood length of the second clustering method when the first and second clustering methods include the DBSCAN algorithm.

10. Method according to any of the preceding claims, wherein the traffic light objects (nc) are recorded on a crossing-specific basis.

11. Device for processing data about a traffic light system (A) that controls a road portion and has at least one traffic light (tp1, tp2, tp3, tp4, tps), wherein the device is designed and configured to carry out a method according to any of claims 1 to 10.

12. Computer program for processing data about a traffic light system (A) that controls a road portion and has at least one traffic light (tp1, tp2, tp3, tp4, tps), including instructions which, when the program is carried out by a computer, prompt the computer to carry out the method according to any of claims 1 to 10.