EP2116981A2 - Method and device for calculating backlog lengths in light signal assemblies - Google Patents

Abstract

The method involves positioning a traffic model i.e. Nagel-Schreckenberg-model, for a road segment with a light signal system, where the traffic model provides minimum density profile in a connection of parameter. Position data is detected from a registered vehicle at the road segment. An evaluating process is carried out for determining the density profile that includes a maximum consistency with the determined position data. A tailback length at the light signal system is determined by a traffic model under the provision of parameter of the selected density profile. An independent claim is also included for a device for determining a tailback length at a light signal system, comprising a data processing device.

Description

The invention relates to a method and a device for determining tailback lengths of traffic signal systems.

The essential problem of traffic situation detection in traffic today is generally based on a comparatively thin data basis to make meaningful or correct statements about the current traffic condition in each considered transport network. The reason for the thin database is primarily to mention that the majority of currently used measuring equipment (induction loops, ...) provides only local information, so that no direct traffic data can be measured for the road sections between two such detectors. Newer types of detectors, such as video cameras or FCD systems, which are in principle capable of a comprehensive detection of traffic, each have other disadvantages.

Overall, additional information must therefore be obtained in addition to the acquired measured values with regard to nationwide traffic situation detection through skilful processing of the respective traffic data, including generally more or less complex traffic models (queue models, cellular automata,...) Or empirical relationships between the data of different detectors. Keyword: data fusion) can be used.

As part of the traffic situation detection, there are various ways of measuring traffic data. Especially with regard to the determination of tailback lengths of traffic lights but - as mentioned above - usually additional considerations necessary because jam lengths can generally be measured directly only with great technical or financial effort.

The simplest method of traffic monitoring is manual traffic counting, where on-site traffic observations are carried out by appropriate persons. This form of traffic situation detection can, however, usefully take place only within a limited time frame and sometimes provides only very rough data. Consequently, such data can hardly be used to obtain comprehensive and timely data Traffic information is used. Rather, they serve as purely historical information that is used primarily for offline planning purposes and as empirical values.

A simple further development is the observation of the road traffic at selected road sections and junctions with the help of video cameras, the images of which are displayed up-to-the-minute on different monitors of a traffic control center. As a result, a relatively large area can be observed by a single person at the same time. Nonetheless, the problem remains that such a system can not be completely automated so that, in addition, the quality depends substantially on the attention and experience of the individual who needs to evaluate the camera images in the traffic control center.

Accordingly, numerous methods have been developed heretofore for fully automatic traffic information, i. Specifically detention lengths at traffic lights, to capture and process. As part of the traffic situation detection with classic loop detectors (induction loops) usually traffic levels, occupancy times of the respective detector and time gaps between the vehicles are measured (depending on the design also local speeds of the vehicles are measurable.). About the occupancy times in a loop detector in the area in front of a traffic signal system can be decided whether a corresponding traffic light backlog is limited to the distance between the traffic signal and detector or at least reaches up to the detector. However, a more exact indication of the backpressure length is not possible with this method.

Furthermore, it should be noted that the validity of the measured values of loop detectors, in particular in urban traffic networks, is generally strongly limited locally. In other words, it is not unlikely that a detector that is perhaps only 100 meters away from another will measure completely different data. Consequently, for immediate, area-wide traffic situation detection with induction loops (the same is true for detectors with similar characteristics, such as infrared or radar sensors) corresponding detectors must be installed in relatively close distances to each other, which is not realistic for the traffic network of an entire city from an economic point of view is.

As mentioned above, therefore, the smarter traffic detection methods either use appropriate traffic models or skillfully link the readings of multiple detectors to obtain more information. Thus, a method with a simple balancing approach is known, in which the backpressure length is estimated by comparing the number of inflowing and outflowing vehicles, the inflow being controlled in each case by means of a detector. In a first version of the method, a simple traffic model is used by assuming a uniform approach of the jammed vehicles with a constant, temporal distance for the departing vehicles. In a second variant, the outflow is also controlled by means of a loop detector.

In both cases, however, jams that extend beyond the detector in front of the traffic light can not be quantified in terms of their length. Furthermore, the method makes certain demands on the spatial position of the induction loops, which is not given everywhere in reality. Finally, if the balancing in the preferred variant takes place with the aid of the comparison of the data of two detectors, this additionally leads to the method reacting very sensitively to erroneous data or the failure of one of the detectors.

Furthermore, a model-based method for traffic position detection on traffic signal systems is known, which (only) requires the counting and speed data of a single loop detector as an input variable on each network edge, but with regard to its relative position to the traffic signal system additional conditions apply. Since the actual distances between the traffic signal system and the detector in many of the already existing counting loops should not in many cases correspond to the required 100 to 150 meters, the widespread application of the method would require the installation of many additional detectors, which does not seem realistic for reasons of cost ,

In order to circumvent the problem of local detection (for example in induction loops), in which data can only be measured at individual road cross sections bordered on a few meters, digital cameras for traffic monitoring have been used for some time now, as already mentioned above Help each an entire area of the transport network (namely, the entire field of view of the camera), such as an intersection, can be detected simultaneously. With the aim of automating the analysis The camera images are followed by a digital image processing that recognizes vehicles (or more generally moving objects) and generates traffic data such as tailback lengths at traffic light systems.

In this context, the correct detection of vehicles due to various effects such as shadows, occlusion or darkness in the recorded camera images is still difficult. In fact, because of the same phenomena, it is sometimes not easy even for a human being to properly distinguish vehicles in the images. Furthermore, it goes without saying that jams can of course only be detected within the field of view of the camera, so that there is also a conceptual upper limit for the detectable length of the traffic light jam here (for example in the case of a camera installed at an intersection).

An equally unproblematic aid in this respect is the comparatively young FCD technology, in which (depending on the design of the system, also completely anonymous) the positions (possibly also other current vehicle data such as speed, turn signal activity, windscreen wiper setting, ...) are like that Floating Cars are sent at certain times to a central office, where they can be evaluated together with the data of other floating cars. Furthermore, since floating cars can be everywhere on the road in general, FCD systems are even stronger than the traffic observation with video cameras and unlike the classic traffic detectors already conceptually to a comprehensive traffic situation detection in a position.

At the same time, however, this also shows the essential weakness of all previous FCD applications: In principle, the FCD fleet sizes used in the current FCD systems are very small in relation to the size of the considered transport network, so that all pure FCD methods are currently used with an extremely high FCD thin data basis. As a result, floating-car data is purely supportive in many applications and merely serves to control traffic data obtained by other means.

In order to be able to determine a traffic situation in the event of missing or too less current floating car data, it is of course possible to store all measured values continuously in a database, so that for each future Date of missing or too little data historical information can be added in a suitable manner to the current database. Depending on the proportion of supplemented historical measured values, part of the timeliness of the traffic condition determined in each case on the basis of the extended data basis is lost. However, as road traffic in general often reveals periodically recurring patterns (eg typical rush hours, ...), a skilful selection of additional historical data (eg taking into account the day of the week and the time of day) can substantially avoid grossly falsifying the results.

With regard to the concrete methods for traffic monitoring using FCD (especially in urban transport networks) are used in most current applications floating cars, in addition to timestamp and current position (possibly also current speed, ...) for each FCD message to the Send each central office a unique vehicle identification number. In this way, the route traveled (trajectory) in the road network can be reconstructed as far as possible by means of suitable map matching and routing algorithms for each vehicle. Furthermore, in each case for the distance between each two consecutive FCD messages from the difference of the associated time stamp results in a travel time, which can be regarded as essential traffic information. In addition, with the help of easily ascertained knowledge about the length of each of these sections (partial trajectories), which may well comprise several network edges, it is possible to directly calculate an average cruising speed for the respective road section being traveled.

Unfortunately, due to the complex dynamics of the traffic flow, the traffic data obtained in this way is very noisy, especially in urban traffic networks, which often complicates the further evaluation. This is because the travel times vary greatly from the particular traffic signal phase that the particular FCD vehicle encounters (e.g., green wave or, conversely, many red phases).

Furthermore, travel times, which in the situation described are generally not necessarily related to the individual network edges of the underlying digital map, are difficult to interpret in the context of traffic management, ie especially with regard to the control of traffic lights.

From the DE 100 18 562 C1 is a method for obtaining traffic data for a traffic network with traffic-controlled network nodes and these connecting route edges by moving mitbewegenden signaling vehicles known, data acquisition operations are not triggered at least for successively traveling network nodes each before leaving a branching into the respective network node route edge and in the respective data acquisition process time stamp information is obtained as traffic data indicating a reporting time related to the respective network node not earlier than the time of leaving the relevant route edge and not later than the time at which the reporting vehicle considered a portion of a subsequently traveled route edge before a next one Network node reached.

From the DE 100 22 812 A1 A method is known for determining the traffic situation on the basis of traffic data obtained by traffic-moving registration vehicles, for a traffic network with traffic-regulated network nodes and connecting route edges, with indicative traffic data for the travel time on the route edges by traffic-moving registration vehicles be obtained, based on the traffic data obtained, the travel times for the route edges are determined and determined based on the determined route-specific travel times one or more traffic situation parameters.

In this case, the required form of the reporting times (message at the network node) increased demands on the equipment of floating cars used compared to the currently commonly used method that floating cars simply at regular intervals (typically 30 seconds to 5 minutes) transmit their current position data ,

Basically, it would be very useful for the evaluation of floating car data, of course, if the recorded vehicle data with the shortest possible time intervals between the individual FCD messages would be available. The main bottleneck of all FCD systems, however, is the communication between the vehicles and the control center. Since the connection is currently usually made via paid mobile networks (GSM, UMTS), a continuous data transfer is not possible even from an economic point of view.

From the DE 10 2004 039 854 A1 a method for determining traffic information in a road network with at least one intersection is known, on which the traffic flow is controlled by a traffic signal having a control unit for switching on light-emitting signaling devices, which correlates with the traffic flow traffic data and from this traffic information, in particular a Traffic demand for at least a portion of the road network are determined, being detected by the road intersection approaching vehicles a sampling fleet vehicle-specific traffic data by locally limited to the intersection, wireless transmission from the vehicle to the control unit of the traffic signal.

From the DE 101 10 326 A1 a method for determining a current traffic situation for traffic constructions and / or traffic forecasts in a traffic network is known, wherein the current traffic situation is determined for a given area based on a location of mobile phones, wherein the location of the mobile phones performed at different successive times and from the determined Locating the mobile phones, a spatial distribution of the mobile phones at the different successive times is determined and stored.

From the DE 101 10 327 A1 a method for determining the current traffic situation for traffic constructions and / or traffic forecasts in a traffic network is known, wherein the current traffic situation is determined for a given area based on a location of the mobile phone, each mobile phone is uniquely identifiable via an associated identification number, the location the identified mobile phones are performed at different consecutive times and determined and stored from the determined locations of the identified mobile phones at the different successive times.

By analyzing the distribution of the mobile phones and in the event that an identification of the individual mobile phones, ie a tracking over time is possible by the evaluation of the movement pattern of each mobile phone can be decided whether a detected mobile phone is moved about in a car belongs to a cyclist or to a pedestrian. Assuming, for example, that in the long term, presumably almost every car driver will carry a mobile phone with him, virtually every vehicle in the road can be compared to his Position are recorded. Subsequently, for example, all the car positions determined in this way can be represented graphically as points in a digital map by means of map matching. Due to a higher traffic density, congestion can easily be identified by the accumulation of such points. The graphics are evaluated either manually or by means of automatic image processing by comparison with previously stored traffic patterns for which the desired traffic situation parameters such as tailback periods have already been determined. However, the method essentially depends on the fact that the largest possible proportion of road users can be detected, so that, regardless of the immense volume of data that would be generated by a widespread use of the method, a possibly desirable transfer of the described graphical method to a classic FCD system, where the tracking accuracy is typically significantly higher, currently seems not possible.

From the article " Nagel, K., Schreckenberg, M .: A cellular automaton model for freeway traffic, I. Phys. France 2 (1992), pages 2221-2229 "is a traffic model known, which is known in the art as a Nagel-Schreckenberg model, for example, the density profiles can be calculated for a density profile is a course of local traffic density on the considered road segment.

The invention is based on the technical problem of providing a method and a device for determining tailback lengths of traffic signal systems, which can determine backstop lengths over a wide range with little metrological outlay.

The solution of the technical problem results from the objects with the features of claims 1 and 9. Further advantageous embodiments of the invention will become apparent from the dependent claims.

The method and the device for determining tailback lengths at traffic light installations by means of a data processing device are characterized in that a traffic model for road segments with traffic signal systems is implemented in the data processing device, the traffic model providing at least density profiles as a function of a parameter, wherein position data of Reporting vehicles are supplied in each road segment and an estimation method for the determination the density profile can be carried out, which has the greatest agreement with the determined position data, wherein a reset length is determined at the traffic signals by means of the traffic model taking into account the parameter of the selected density profile. The position data can be highly accurate GPS data, but also, for example, position data that has been determined by means of a mobile telephone. In this case, an identification of the reporting vehicles and / or mobile phones is not mandatory. Furthermore, the method requires only a few position data to deliver useful results. The position data are preferably provided with a time stamp before they are transmitted to the device. It should be noted that the term "traffic model" generally also means a suitable, parameter-dependent, mathematical function.

Possible applications of the backlog length data thus determined are e.g. Quality assurance in traffic management, for example, in the context of controlling the effects of changes in the traffic light circuit diagrams or traffic-dependent navigation in urban road networks. In the case of an adequate supply of floating-car data, an online traffic management in the sense of an up-to-the-minute, traffic-dependent traffic influencing (traffic signal control, dynamic routing, etc.) is also conceivable.

Moreover, the method according to the invention is naturally also suitable for all other forms of traffic or transport networks with similar framework conditions (edges with periodically controlled outflow) in order to detect congestion of the respective traffic objects at the edge end with respect to their length.

Moreover, the method implicitly incorporates some mechanism for self-correction, so that even the few required parameters compared to other approaches of the model-based traffic situation detection often need only be roughly estimated. As in the case of the database, the method according to the invention therefore places comparatively small demands on the required model parameters, which on the one hand minimizes the calibration effort and on the other hand additionally makes the method more robust in a certain way.

In a preferred embodiment, the parameters segment length (road length) and control parameters of the traffic signal system are supplied to the traffic model, so for example the duration of the red and green phases.

Further preferably, the traffic model is additionally supplied with a maximum speed of the motor vehicles and / or a correction term as a parameter. The maximum speed may be the legal maximum speed or the empirically determined actually driven maximum speed, the latter being preferred.

In a further preferred embodiment, the density profiles depend on the parameter inflow or inflow probability / traffic demand.

In a further preferred embodiment, the traffic model is designed as a Nagel-Schreckberg model. The advantage of this model is that it is not too complicated and yet provides sufficiently accurate density profiles. Furthermore, the model not only provides mean or maximum values for the respective desired traffic parameters, but even complete, approximate probability distributions for the corresponding parameters, in particular the tailback lengths at LSAs.

In a further preferred embodiment, the estimation method is a maximum likelihood estimation. The method of the maximum likelihood estimation has the advantage that the assignment of the corresponding reference is less dependent on a subjective impression of the observer and can be carried out with the aid of a standard numerical optimization algorithm, which sometimes additionally provides a quality measure for the reliability or unambiguity of the Can specify assignment. Another advantage is that very simple additional data (keyword data fusion) in addition to the position data in the determination of the most appropriate density profile can be considered, which in particular with only a small database of position data still a very good estimate can be achieved (restriction of the parameter space by the additional Dates). Instead of limiting the parameter space, individual parameters can also be weighted on the basis of a priori information.

In a further preferred embodiment, in addition to the position data, further data of the reporting vehicles (eg speed) and / or of external sensors (eg induction loops) and / or a priori information (eg historical profiles of the density profiles as a function of day of the week and time) are included in the estimate Determination of the density profile.

In a further preferred embodiment, the traffic model determines further traffic parameters of the road segment and / or generates parameters for adaptive control of the traffic signal system.

The method described above for the determination of tailback lengths at traffic light systems is generally usable for determining tailback periods at temporarily blocked outputs of edges of a transport network. In this case, a model is used to obtain density profiles as a function of a parameter, whereby the model can also consist of simple mathematical functions. The objects moving on the transport network then output position data, in which case the density profile which best matches the position data is again determined by means of an estimation method, and the associated parameter or parameters are fed back into the model and from this a backlog length is calculated. A possible application in logistics is the tracking of packages, containers or similar objects, which are partly designed with means for transmitting position data. The transport network can be designed, for example, as a conveyor belt or conveyor belt. If, for example, the conveyor belt is then stopped at one point, because, for example, there is currently no truck or similar means of transport in order to transport the objects further, a backlog occurs. On the basis of the calculation of the backwater, countermeasures can be initiated or transport times can be better predicted.

Fig. 1

ein schematisches Ablaufdiagramm eines Verfahrens zur Ermittlung einer Rückstaulänge an einer Lichtsignalanlage,

Fig. 2

Verkehrsdichteprofile für verschiedene Zufluss-Wahrscheinlichkeiten,

Fig. 3

eine Zuordnung eines Dichteprofils zu gemessenen Positionsdaten,

Fig. 4

ein Verlauf der Durchschnittsgeschwindigkeit über der Tageszeit,

Fig. 5

einen Wochengang der maximalen Staulänge,

Fig. 6

einen geglätteten Tagesgang der Staulänge,

Fig. 7

ein FCD-Dichteprofil,

Fig. 8

einen Tagesgang der Staulänge und

Fig. 9

einen geglätteten Tagesgang der Staulänge.

The invention will be explained in more detail below with reference to a preferred embodiment. The figures show:

Fig. 1

a schematic flow diagram of a method for determining a Rückstaulänge at a traffic signal system,

Fig. 2

Traffic density profiles for different inflow probabilities,

Fig. 3

an assignment of a density profile to measured position data,

Fig. 4

a course of the average speed over the time of day,

Fig. 5

a weekly cycle of maximum stowage length,

Fig. 6

a smoothed daily routine of the stowage length,

Fig. 7

a FCD density profile,

Fig. 8

a daily routine of the stowage length and

Fig. 9

a smoothed daily routine of the stub length.

The original input variable of the method according to the invention form the position data x _{i of} a sample of relevant traffic objects (floating cars or detector vehicles), which are present as a result of map matching network edge related as distances to any, but fixed reference point of the respective road section. As a reference point, for example, the segment start or the stop line of the traffic light system can be selected at the end of the segment. Depending on the data situation (number of currently recorded FCD positions), the database can also be supplemented to any extent with historical data from a database. In particular, in an advantageous embodiment of the method according to the invention, assuming periodically recurring traffic patterns, position data of the same day of the week and / or the same time of day stored after network edges can be differentiated to the input variables of the method if the FCD coverage is too low. As already mentioned above, it is completely irrelevant in which way the required position data are / were recorded. For example, with suitable pre-processing, ie differentiation according to types (pedestrians, cyclists, vehicles, etc.), the position determination of the relevant traffic objects by means of mobile telephones is possible.

Fundamental to the method according to the invention is now the fact that within an (urban) road network, it is typical for traffic light installations to accumulate vehicles and wait. You go Now from a (nearly) homogeneous distribution of the available floating cars on the amount of all vehicles involved in traffic, it can be assumed that the floating cars are spatially scattered essentially according to the current profiles of local traffic densities in the transport network. In other words, in sections with a high local traffic density (eg, traffic light systems), there are relatively more floating cars than, for example, areas with free traffic flow and correspondingly lower local traffic density.

On the basis of a traffic model, which represents one of the essential components of the method level of the method according to the invention, corresponding profiles of local traffic densities can then be derived analytically or by simulation. In an advantageous embodiment, which supplies the method to a more efficient numerical feasibility, the determination of the required density profiles K ( q ) takes place within the framework of a mathematical analysis. The density profiles are determined as a function of a certain inflow probability q , which essentially corresponds to the traffic demand. A selection of such profiles for different q shows Fig. 2 , where very well the significantly higher local traffic density in the area of the traffic signal system (right edge of the graph) and the traffic density increasing overall with increasing traffic demand / inflow probability q are recognizable.

The required parameters of the Nagel-Schreckberg model, which can vary depending on the network edge, are the respective length L of the relevant road section (typically segment start to stop line), the (maximum permissible) speed v _max and the traffic light phase g for the (effective ) Green and r for the (effective) red phase. Yellow phases are neglected in the concrete embodiment of the method according to the invention, but can be taken into account as needed.

Under the condition applicable to many current FCD systems, the available floating cars now transmit their respective current position independent of the surrounding traffic situation and without grid reference in the reporting strategy, ie, for example, at regular time intervals or initially store it internally within the framework of a suitable system architecture and later send as a package, but then not only the Vehicles themselves, but also the reported FCD positions distributed according to the current density profile of the associated road section. In other words, if in the data acquisition process used there are no special event or network-related reporting times (or times for temporary in-vehicle storage) of the current FCD positions, FCD messages occur on the respective road segment in the area of the traffic light backlog (quantitatively the associated FCD positions Density profile correspondingly) is more likely to occur than on the part with free traffic.

Assuming now for the (in principle arbitrary) period during which the FCD positions used as the data basis were registered, a stationary traffic situation, ie assuming that all FCD positions according to the same density profile K ( q *) with a unique assigned , but unknown q * ∈ [0,1] are distributed, one has to ask for a statistical estimate of q * for which density profile or traffic situation the combination of all position data has the maximum probability.

This is the essential part of the process module "Fusion" in Fig. 1 , By means of a statistical standard method, the so-called maximum likelihood estimation, the q * in the so-called parameter space Θ: = [0,1] is determined, whose unambiguously assigned density profile K ( q *) is determined by all the density profiles K ( q ) with q ∈ Θ fits in some way best to the observed FCD positions. Fig. 3 shows by way of example the assignment of a density profile K ( q *) to some real measured position data in the form of a frequency profile (The individual columns indicate how often within the observation period at the corresponding position (meter-accurate) of the road section registers a floating car was.) Are shown.

In the context of the maximum likelihood estimation, it may be advantageous for technical reasons to first limit the initially uncountable parameter space Θ and thus to reduce the amount of possible traffic conditions, more precisely reference density profiles K ( q ) in advance, or more generally in advance the estimation of the parameter q * an a priori weighting of the permissible parameters make. It is conceivable, for example, to consider only the integer multiples of a certain step size h instead of the entire, real interval [0,1], so that the restricted (now finite) parameter space is given by Θ: = { q ∈ Θ | q = k · h, k ∈ IN ₀ }.

At the same time, this possibility of being able to manipulate the parametric parameter space, which is fundamental for the maximum likelihood estimation, virtually in the manner of an interface provides an interface for integrating additional traffic information into the method. If, for example, current speed data v _{i of} the relevant floating cars are available and, on average, comparatively low, this may indicate a rather high traffic demand (cf. Fig. 4 : In the speed aisle, the break-ins during peak hours in the morning and in the late afternoon can be clearly seen.), So it makes sense to exclude small values for q * directly or at least with a smaller weight. Conversely, if, for example, a loop _detector at the entrance of the considered road section yields very low traffic intensities q _detector , it can generally be assumed that the demand for traffic will be low and, in principle, rather small q *.

If, due to other traffic data, an unambiguous traffic situation with the corresponding q * is already known, the parameter space Θ can even be formally reduced to a single value, namely q *, in a highly flexible manner. In other words, in the context of the method according to the invention, the result of the maximum likelihood estimation described would already be clearly defined even before it is used.

Another important advantage of the statistical method used in the method according to the invention is also a special form of self-correction is that regardless of the actual quantitative correctness of the reference density profiles K ( q ) that is always selected for describing the traffic situation, which in certain Best fits the observed FCD positions. In particular, this is the case if the model parameters (especially v _max , g and r ) do not correspond exactly to the real conditions. Because of the basically qualitative similarity of Density profiles K ( q ) for different constellations of the model parameters can, however, to a certain extent be determined even if the model is incorrectly calibrated, an approximately correct traffic situation (self-correction), ie in particular the correct tailback length at traffic light installations. Only the corresponding estimate for q *, which in some way represents a kind of correction term, should in this case generally no longer match the actual traffic demand, even though q * is the input of the final result generation in the lower "Model" block (see Fig. 1 ) is still of crucial and beneficial importance.

The calculation of the desired traffic situation parameters is in fact in complete consistency with the rest of the inventive method by a return of the estimated q * in the traffic model used, from the original, analytical determination of the multiple reference density profiles K ( q ) directly concrete formulas such as the expected, average tailback length L _{congestion, average} ( q *), the expected, maximum stowage length L _{congestion, max} ( q *) and their standard deviations σ ( L _{congestion, average} ( q *)) and σ ( L _{congestion, max} ( q *)) can be derived. Moreover, of course, the estimated value q * and the associated density profile K ( q *) can be output.

Finally, a final peculiarity, which additionally distinguishes the method according to the invention against other approaches to traffic position detection, is the fact that in addition to all these concrete values, even a complete probability distribution π ( q *) derived for the model analysis for the number of vehicles in the respective road section at the end of a red phase can be specified, which represents in a special way at the same time, among other things, a distribution for the average and maximum backwater length. Consequently, in addition to the aforementioned traffic parameters, all values (eg variance, higher moments, quantiles) which are contained as distribution properties in π ( q *) can be determined.

Finally, in order to check the mode of operation of the method according to the invention and to prove the advantageousness, an evaluation was carried out on the basis of real traffic data (position and speed data). By Summary of all FCD positions of a month was differentiated by weekdays with a data density like in Fig. 3 in the Fig. 5 shown week-long line for the maximum tailback length at the traffic signal system of the considered road segment. You can see a typically expected course of the average shortest backlogs on the weekend, especially on Sunday.

With regard to the detection of daytime fluctuations in the tailback length, an additional data set with real position and speed data of floating cars was evaluated for another road section over five weeks. The data was grouped hourly by time and then summarized on all working days (Monday to Friday) away. As a result, there were 24 individual data records (one for each hour of the day) that contained the position data of all working days of the data collection period for each hour and were then evaluated individually. The resulting diurnal line of a typical working day for the tailback length at the traffic light system under consideration is in a minimally smoothed form in FIG Fig. 6 shown. You can clearly see the two rush hours in the morning and in the afternoon with at the same time very short traffic jams, especially at night.

It is noteworthy that by differentiating the FCD position data after the hours of the day, the individual hourly data records in comparison to the previous example (data density: over 400 data points) form an extremely thin database (cf. Fig. 7 : The less than 30 data points correspond to an average of just about 5 registered floating cars on the considered road section per hour.), With which the method according to the invention still seems to be doing well.

In fact, in other cases, the situation sometimes becomes more difficult (s. Fig. 8 The method according to the invention can hardly distinguish times with very high and very low traffic volumes in the present low database.), And it shows up a little more plausible depending on the specific data situation (at high data densities this phenomenon should generally not occur) Daily routine for the backstops.

In the concrete example, however, the available FCD speed data could additionally be made available via the data fusion interface described above, in that the parameter space Θ of the maximum likelihood estimation in the context of the method according to the invention is based on small or large values of the traffic demand / Inflow probability q was limited, with a complete supersaturation of the considered road section was excluded in principle. The significantly better and again completely plausible, minimally smoothed results shows Fig. 9 , while emphasizing that the same database as in Fig. 8 underlying. Overall, it was thus possible to demonstrate in an advantageous manner that good results can still be achieved with the method according to the invention even with a comparatively thin or apparently insufficient database.

• Bisher wurde im verwendeten Verkehrsmodell keine spezielle Unterscheidung nach Richtungsspuren vorgenommen. Insbesondere können dadurch zunächst keine Rückstaulängen etwa für die unterschiedlichen Fahrtrichtungen (rechts, links, geradeaus) an Kreuzungen unterschieden werden. Ist aus den verwendeten Floating Car Daten allerdings entweder aufgrund einer hochgenauen Ortung oder z.B. durch Kenntnis der Fahrtrouten eine Zuordnung der einzelnen Fahrzeuge zu jeweils einer Richtungsspur möglich, so kann in einfacher Weise durch Gruppierung des Datensatzes nach diesem Kriterium leicht eine richtungsspurbezogene Auswertung mit Hilfe des erfindungsgemäßen Verfahrens realisiert werden.
• Wie bereits mehrfach erwähnt ist für das erfindungsgemäße Verfahren grundsätzlich eine Kopplung mit einer Systemarchitektur wie beispielsweise in DE 10 2004 039 854 A1 beschrieben möglich. Dadurch können wie dort behauptet zum einen die Kosten zur Kommunikation zwischen Fahrzeugen und Zentrale bei gleichzeitig starker Erhöhung der Datendichte drastisch reduziert werden. Folglich können in diesem Kontext - wie überhaupt bei jeder Form der Verbesserung der Datengrundlage, z.B. mittels Mobilfunktelefon-Ortung wie in der DE 101 10 326 A1 und DE 101 10 327 A1 - mit dem erfindungsgemäßen Verfahren noch deutlich genauere und aktuellere Verkehrsinformationen erfasst werden, so dass unter Umständen insbesondere auch Online-Verkehrsmanagement-Anwendungen möglich werden.
• Wegen seiner hohen Flexibilität ist das erfindungsgemäße Verfahren sowohl mit klassischen FCD-Systemen (direkte Kommunikation zwischen Floating Cars und Zentrale) als auch mit nahezu jeder Form der neueren (teilweise dezentralen) Ansätze (Car-2-Infrastructure und Car-2-car) kompatibel. Folglich ist eine vielseitige und langfristige Einsetzbarkeit des erfindungsgemäßen Verfahrens grundsätzlich vorstellbar.
• In seiner beschriebenen Ausführung für den Straßenverkehr ist das erfindungsgemäße Verfahren ein eindimensionaler Ansatz in dem Sinne, dass die Verkehrsdichte entlang einer (eindimensionalen) Strecke (Straßenabschnitt) betrachtet wird. Formal ist mit den verwendeten mathematischen Methoden aber auch jede höher-dimensionale Variante durchführbar. Insbesondere könnte etwa auch der Verkehr auf einer zweidimensionalen Fläche mit den gleichen Methoden erfasst werden, wobei in diesem Fall die Referenz-Dichteprofile anschaulich zu Dichtegebirgen werden.
• Entsprechend ist die Anwendung des erfindungsgemäßen Verfahrens konzeptionell nicht auf Straßenverkehrsnetze beschränkt, sondern kann grundsätzlich auf beliebige Verkehrs- und Transportnetze wie z.B. den Fußgängerverkehr (zweidimensional) oder den Luftverkehr (dreidimensional) übertragen werden.
• Wie bei jedem Verfahren zur Verkehrslageerfassung können überdies die Ergebnisse kontinuierlich in einer Datenbank gespeichert und zu verschiedenen Zwecken weiterverwendet werden. Zu nennen sind hier beispielsweise strategische Verkehrsmanagement-Anwendungen (Planung, Qualitätssicherung, ...) oder Ansätze zur Verkehrsprognose auf Basis historischer und/oder aktueller Verkehrsdaten.

Finally, of course, general expansion options should be mentioned, which sometimes include other fields of application:

• So far, no particular distinction was made in the traffic model used for directional tracks. In particular, this means that initially no tailback lengths can be distinguished, for example, for the different directions of travel (right, left, straight ahead) at intersections. Is from the floating car data, however, either due to a highly accurate location or eg by knowing the routes an assignment of each vehicle to each direction track possible, it can easily by grouping the record according to this criterion easily a direction-track-related evaluation using the invention Procedure be realized.
• As already mentioned several times is for the inventive method in principle a coupling with a system architecture such as in DE 10 2004 039 854 A1 described possible. As a result, as claimed there, the costs for communication between vehicles and the central office can be drastically reduced while at the same time the data density is increased significantly. Consequently, in this context - as in any form of improvement of the data base, for example by means of mobile phone positioning as in the DE 101 10 326 A1 and DE 101 10 327 A1 - With the inventive method even more accurate and up to date Traffic information are recorded, so that under certain circumstances, especially online traffic management applications are possible.
• Because of its high flexibility, the method according to the invention is compatible both with classical FCD systems (direct communication between floating cars and central office) and with virtually every form of the more recent (partially decentralized) approaches (Car-2 Infrastructure and Car-2-car) , Consequently, a versatile and long-term applicability of the method according to the invention is basically conceivable.
• In its described embodiment for road traffic, the method according to the invention is a one-dimensional approach in the sense that the traffic density along a (one-dimensional) route (road section) is considered. Formally, with the mathematical methods used, however, any higher-dimensional variant can be carried out. In particular, for example, the traffic on a two-dimensional surface could also be detected by the same methods, in which case the reference density profiles become clear to density mountains.
• Accordingly, the application of the method according to the invention is conceptually not limited to road transport networks, but can in principle be transferred to any traffic and transport networks such as pedestrian traffic (two-dimensional) or air traffic (three-dimensional).
• In addition, as with any traffic assessment procedure, the results can be continuously stored in a database and used for various purposes. These include, for example, strategic traffic management applications (planning, quality assurance, ...) or approaches to traffic prognosis based on historical and / or current traffic data.

Claims (16)

Method for determining tailback lengths at traffic light installations by means of a data processing device, comprising the following method steps: a) establishing a traffic model for road segments with traffic signal systems, the traffic model providing at least density profiles as a function of a parameter, b) detecting position data of reporting vehicles in the respective road segment, c) performing an estimation method for determining the density profile, which has the greatest agreement with the determined position data, and d) determining a tailback length at the traffic signals by means of the traffic model, taking into account the parameter of the selected density profile. A method according to claim 1, characterized in that the traffic model, the parameters segment length and control parameters of the traffic signal system when setting up the traffic model are supplied. A method according to claim 2, characterized in that the traffic model additionally a maximum speed and / or a correction term are supplied as parameters. Method according to one of the preceding claims, characterized in that the density profiles are dependent on the parameter inflow. Method according to one of the preceding claims, characterized in that the traffic model is designed as a Nagel-Schreckberg model. Method according to one of the preceding claims, characterized in that the estimation method is a maximum likelihood estimate. Method according to one of the preceding claims, characterized in that in addition to the position data further data of the reporting vehicles and / or of external sensors and / or a priori information in the estimate to determine the density profile. Method according to one of the preceding claims, characterized in that further traffic parameters of the road segment are determined by the traffic model and / or parameters for an adaptive control of the traffic signal system are generated. Device for determining tailback lengths of traffic signal systems, comprising a data processing device, wherein in the data processing device a traffic model for road segments is implemented with traffic signals, wherein the traffic model at least density profile as a function of a parameter provides, being supplied via an interface of the data processing device position data of reporting vehicles in each road segment and an estimation method for determining the density profile is feasible, which has the greatest agreement with the determined position data, wherein a backstop length at the traffic signal installations is determined by means of the traffic model taking into account the parameter of the selected density profile. Apparatus according to claim 9, characterized in that the traffic model, the parameters segment length and control parameters of the traffic signal system are supplied. Apparatus according to claim 10, characterized in that the traffic model additionally a maximum speed and / or a correction term are supplied as parameters. Device according to one of claims 9 to 11, characterized in that the density profiles are dependent on the parameter inflow. Device according to one of claims 9 to 12, characterized in that the traffic model is designed as a Nagel-Schreckberg model. Device according to one of claims 9 to 13, characterized in that the estimation method is a maximum likelihood estimate. Device according to one of claims 9 to 14, characterized in that in addition to the position data received further data of the reporting vehicles and / or external sensors and / or a priori information in the estimate for determining the density profile. Device according to one of claims 9 to 15, characterized in that the traffic model determines further traffic parameters of the road segment and / or the data processing device generates parameters for adaptive control of the traffic signal system.

Images