EP1057156B1 - Detection of traffic situation with fuzzy classification, multi-dimensional morphological filtration of data and dynamic construction of domains - Google Patents

Abstract

The present invention relates to a method for generating road information that indicates the traffic situation of a road network, wherein said method comprises processing traffic measurement values acquired at different moments. The method of the present invention comprises inputting and storing into archive windows the traffic measurement values acquired for each observed street according to the place (x) and time (t) of their acquisition and into categories of measurement values, wherein said windows are continuously actualised, cover a precise period of time from the current moment of road information generation towards the past, and discretise the time and place into intervals. This method is characterised in that the traffic measurement values acquired in the different archive windows observed are filtered using different filters as well as the time and place curve thereof, a characteristic being generated for each filter. The different characteristics are then grouped in order to obtain a characteristic vector related to each place in the road network and describing the traffic situation. The method further includes generating road information which can be transmitted and which is derived from the characteristic vectors describing the local traffic situation.

Description

Background of the Invention

in traffic, especially on motorways and expressways the traffic conditions change very quickly. As a result of the high flow of traffic and the individual driving style of the road users creates a dynamic in Traffic that can hardly be identified with foresight. Add to that Interference influences such as speed limits, changing the number of Lanes, construction sites and accidents that suddenly change to one lane or the whole Block the freeway. The disturbances then spread out like waves and also lead at greater distances to impair traffic. Also the Oncoming lanes are usually affected because road users there too given the happening their speed for various reasons curb. This traffic dynamics often leads to individual or serial rear-end collisions, because the safety distance was not kept or the Drivers simply no longer master this critical traffic dynamics.

There is therefore an urgent need for timely and up-to-date information about the traffic conditions in sections of the route that are identified as precisely as possible. Not only is the traffic jam event important, but also the route-related indication of traffic conditions that usually precede a traffic jam, for example stagnant or heavy traffic. A differentiated detection of the traffic conditions is therefore desirable and necessary, despite the frequent lack of additional information about the current status of the routes, e.g. about construction sites, number of lanes or their topological course: inclines, downhill gradients, etc.
The traffic domains with traffic conditions that are the same or similar everywhere in a traffic domain spread, they grow, divide, migrate, and the traffic conditions merge into one another until they finally dissolve, i.e. there is free movement again. These domains must therefore be found, classified, localized and tracked dynamically.

This is followed by the message management for traffic reports, the ultimately the results are prepared in a form suitable for drivers.

The following problems exist in realizing this objective:

The traffic situation cannot be seen from the individual traffic measurements. To the sizes and changes in traffic measurements over a particular Time interval can be considered. This "integration" allows the course thus determined the traffic measurements do not lose their topicality.

It is also necessary to get traffic measurements from different sensors process that can be stationary or mobile, synchronous, asynchronous or supply event-induced measured values and their measured values also incomplete and very may be noisy.

It is also desirable that the process not follow a predetermined one Route allocation is bound, i.e. works independently of route and also Route knowledge about number of lanes, construction sites, speed limits, Topology of the route sections etc. are not required.

From the traffic measurement data it is not possible to clearly determine the traffic situation at the respective measuring points are closed. For this it is necessary to follow the course of the Knowing measured values about place and time. The specific characteristics of traffic are only included in the measured values together with their local-time profile. the goal is hence the summary of traffic data different physical Content to resultant feature vectors, which clearly the traffic situation characterize. In addition, there is also a distinction between the states "no traffic" and "total congestion", which give the same measured values, are required.

The inventive method according to claim 1 solves these problems consistently.

The creation of traffic information according to claim 1 can be immediate or after intermediate steps. So before creating Traffic information in claim 1 intermediate steps according to characteristics of Subclaims are made.

In the further processing of the traffic situation Characteristic vectors result in the following further problems:

The differentiated traffic conditions, e.g. scattered, stagnant, dense and free are not clearly definable and delimitable with regard to the input data. binary Border crossings between these traffic conditions also do not correspond to that subjective feeling of road users. It is therefore a gradual one Description of traffic conditions and a smooth transition from state to Condition necessary.

The description of the traffic conditions in one place must be comparable It is also a uniform scaling for subsequent calculations and the All characteristics required in one place.

The subject of claim 2 is a method that solves these tasks consistently.

The further processing of such a traffic condition description results following problems:

The traffic measurements, which are recorded with different sensors and as Source information used for the traffic condition descriptions are both local as well as patchy. Gaps can also arise from the fact that there are no or too few measurements in one place. In these cases, the Traffic conditions at these locations cannot be determined directly.

The traffic conditions can only be determined directly at the measuring points. There not any number of sensors can be attached to the road network, there are also no traffic measurements between the measuring points.

However, the goal is a continuous description of the traffic conditions throughout the whole Route and ultimately for the entire road network.

This objective cannot be achieved by extrapolating the measured values themselves become. Instead, there is an interpolation of the values between two measuring points complex and the interpolation range must be limited. You can fill in the gaps also fill with averaged historical data. But this leads to falsifications. Substitute values from model calculations to close the measurement value gaps based on the Available measured values are complex and set a high quality for them available measurements ahead.

The subject of claim 3 solves these problems consistently.

For the formation of traffic domains in each case uniform traffic condition as well their tracking over time and place must compare state vectors become. The following problems arise:

The determined traffic situation at the different locations of the Road network must be one of the differentiated traffic conditions (congested, halting, dense or free). For this purpose, the Descriptions of local traffic conditions can be compared.

Likewise, the locations combined into domains must have the same traffic conditions be compared with each other to see changes and dynamics of these domains to be able to determine in terms of place and time.

However, these comparisons cannot be identical based on identity comparison same traffic measurements or the characteristic feature vectors that the describe the local traffic situation in another form, because these stochastic values are never completely the same, from different ones physical measurements and thus have different units of measurement and are scaled differently.

The aim is therefore, based on the location state vectors, which in their components containing the probabilities for the differentiated traffic conditions with the help a similarity measure, both the location state vectors and the domains to compare with each other.

The subject of claim 4 is a method that solves these tasks consistently.

When segmenting traffic domains in road networks, the following problems can be solved:

Places with the same or similar traffic conditions must be the same as domains Traffic conditions are summarized or the domains more similar Traffic conditions must be segmented from each other.

The domains should not be bound to fixed sections of the route, but rather to them The beginning and end should correspond to the actual traffic days and let their dynamics be fixed in the right places.

The domains should contain a description of the future, which is derived from the State descriptions of the locations summarized in them result.

The subject of claim 1 is also a method that accomplishes these tasks consistently solves.

When tracking domains and assigning them to previous ones Domains reported to Vars have the following problems:

The domain segmentation of the road traffic network must be at time intervals be repeated. These update steps must ensure that the Segmentation always corresponds to the current traffic situation.

The new traffic domains currently found must match the existing, domains found in the previous update step.

The message management must use this to create new, change and delete messages produce.

The traffic situation classification and reporting method according to the invention is on Can be used on all motorways and expressways to reduce the risk of accidents. Critical traffic situations can sometimes even be avoided. The system is not only necessary for security reasons. It is also the basis for Route planning, travel time forecasts, traffic jam prognoses, detour recommendations, Signal and traffic influencing as well as direct if necessary Vehicle control and traffic control and numerous other services.

State of the art

Previously known dynamic traffic balancing methods were further modified to avoid typical weaknesses, for example described in the document by Siemens WO9525321A1: "Method of sensing traffic and detecting traffic situations on roads, preferably freeways", 1995 and 1997. This modified method for traffic situation detection is the basis is a measuring section between two measuring points and the formation of the difference, in this case, however, not the traffic flows, but the speed-traffic density values, which are calculated from the local data. In addition, a trend factor is determined from the ratio of traffic flows between the two measuring points over a predetermined period of time, e.g. 30 minutes and the first derivative, i.e. the slope of the tangent, is calculated. Using these fuzzy logic, an indicator of a critical traffic situation in the measurement section is detected from these three traffic parameters
A further development of the same method uses a dynamic calibration of the traffic parameters mentioned, depending on their past values, to correct the difference formation (Siemens / WO9525321A1: "Method of Detecting Traffic and Traffic Situations on Roads, preferably Motorways"). The calibration factor for the speed-density difference is also used as a threshold value, the exceeding of which indicates a critical traffic situation. With these measures, the dependency of the difference formation on many influencing variables such as the distance between the measuring points, the measuring errors and tolerances, the geometry or topology of the measuring section, the road condition, the different driving styles during day and night etc. are corrected.

With these accounting methods, the traffic conditions are classified directly based on the measured values by introducing the measured values into areas. In which The latter method is based on the flow of traffic, the speed and the integrated flow balance a fuzzy logic applied to classify the Traffic conditions that have already been definitively determined. A Different characteristics are not summarized. The so determined Traffic conditions for the respective measuring sections are not continuous, sliding Transitions cannot be adequately described. The real traffic dynamics is largely suppressed when recording the traffic conditions.

Since all of these procedures only with a fixed division of the whole considered transport network, the decision for a particular Traffic condition based on the measured values at the beginning and end of a test section therefore only be taken for the respective route. With these procedures can in principle be the gradual course of the information about the condition under consideration Distance can not be determined.

Accordingly, there are no gradual processes in these processes Grading. That is why they cannot do any on such Condition-based similarity measure for the assignment of the Traffic conditions at the different locations to the differentiated traffic conditions use.

Therefore, only the route sections can be summarized below in which, based on the measured values, the Decision for the same traffic condition was made. These systems can start and end of traffic domains in which the same Traffic conditions prevail, not identify them, but only within the framework of the function-related route division. A dynamic domain formation is with the known methods are not possible in principle.

This means that dynamic domain tracking is not possible. That too Message management is subject to these restrictions.

• Nur synchrone Meßdaten von stationären Sensoren können verarbeitet werden.
• Die Verfahren sind alle streckengebunden, wodurch die örtliche Auflösung eingeschränkt ist.
• Als Merkmal wird nur die über die Zeit integrierte Flußdifferenz berechnet.
• Der Verlauf der Meßwerte über Ort und Zeit wird nur integriert über die festgelegte Meßstrecke, das heißt an zwei Meßpunkten, ermittelt.
• Es gibt keine graduelle Beschreibung des Verkehrszustandes über den Ort, der auch gleitende Übergänge adäquat erfassen kann.
• Eine Ähntichkeitszuotdnung auf Basis einer kontinuierlichen Zustandsbeschreibung ist daher weder für einzelne Orte noch für ausgewiesene Verkehrsdomänen möglich.
• Eine dynamische Bestimmung und Verfolgung der Verkehrsdomänen Ober den Ort und die Zeit ist damit nicht möglich. Das Meldungsmanagement unterliegt den gleichen Einschränkungen

The known methods do not solve the problems presented and they do not offer a solution:

• Only synchronous measurement data from stationary sensors can be processed.
• The procedures are all route-bound, which limits local resolution.
• As a feature, only the flow difference integrated over time is calculated.
• The course of the measured values over time and place is only determined in an integrated manner over the defined measuring section, that is, at two measuring points.
• There is no gradual description of the state of the traffic about the location, which can also adequately record smooth transitions.
• A similarity allocation based on a continuous description of the condition is therefore not possible for individual locations or for designated traffic domains.
• A dynamic determination and tracking of the traffic domains about the location and the time is therefore not possible. Message management is subject to the same restrictions

The known methods are therefore for the practice of detecting the Traffic condition at individual locations as well as the traffic days in Road network can only be used to a limited extent.

Fig. 1:

Zeitfunktionen der Meßwerte von Induktionsschleifen für ein Stauereignis,

Fig. 2:

Mediangefilterte Zeitfunktionen aus Fig. 1,

Fig. 3:

Dilatierte Tophat-Funktion der Breite 15 Minuten angewendet auf die Geschwindigkeitsmeßwerte bei stockendem Verkehrszustand,

Fig. 4:

Schematische Darstellung der Verarbeitungskette:
Eingehende Verkehrsmeßwerte:
Eintrag in Speichereinheiten für die Speicherung von erfassten Verkehrsmesswerten je betrachteter Straße in Abhängigkeit von Ort und Zeit ihrer Erfassung, im folgenden Historienfenster genannt, über Ort x und Zeit t, Bildung von Merkmalsvektoren je Ort nach den orts-zeitlichen Filterungen der Meßwerte,

Fig. 5:

Skizze eines Merkmalsraumes mit Klasseneinteilung.

Fig. 6:

Fuzzy-Klassifikations- bzw. Disknminanzfunktionen für die örtlich und zeitlich gefilterten Meßgrößen (oder die Merkmale),

Fig. 7:

Bestimmung des normierten Ortszustandsvektors an einem Ort x aus den Zustandsvektoren je Merkmal, die sich aus den Merkmalen durch Fuzzy-Klassifikation ergeben,

Fig. 8:

Zustandsextrapolation der Zustandsvektorkomponenten über den Ort mit z.B. einem lokalen Gaußfilter,

Fig. 9:

Skizze zur Visualisierung der Ergebnisse des Domänenwachstumsverfahrens auf Basis der Ortszustandsvektoren,

Fig. 10:

Skizze zur Visualisierung der Ergebnisse und die zeitliche Zuordnungen der durch die Meldungen
ausgewiesenen Verkehrsdomänen zu aufeinanderfolgenden Zeitpunkten,

Fig. 11:

Erzeugte Verkehrsmeldungen des realisierten Prototyps am Beispiel eines Stauereignisses.

Fig. 12:

Erzeugte Verkehrsmeldungen des realisierten Prototyps am Beispiel eines Stauereignisses bei gleichzeitiger Darstellung der Geschwindigkeitsmeßwerte im Hintergrund.

Particular advantages and further features of a method and system according to the invention result from the following description of an embodiment with reference to the drawing. It shows

Fig. 1:

Time functions of the measured values of induction loops for a traffic jam event,

Fig. 2:

Media-filtered time functions from FIG. 1,

Fig. 3:

Dilated tophat function of 15 minutes in width applied to the speed measurements when traffic conditions are slow,

Fig. 4:

Schematic representation of the processing chain:
Incoming traffic measurements:
Entry in storage units for the storage of recorded traffic measured values per street in question depending on the location and time of their acquisition, hereinafter referred to as the history window, via location x and time t, formation of feature vectors per location according to the spatial filtering of the measured values,

Fig. 5:

Sketch of a feature room with classification.

Fig. 6:

Fuzzy classification or discriminance functions for the locally and temporally filtered measured variables (or the characteristics),

Fig. 7:

Determination of the normalized location state vector at a location x from the status vectors per feature, which result from the features by fuzzy classification,

Fig. 8:

State extrapolation of the state vector components over the location using, for example, a local Gaussian filter,

Fig. 9:

Sketch to visualize the results of the domain growth method based on the location state vectors,

Fig. 10:

Sketch to visualize the results and the temporal assignments of the messages
designated traffic domains at successive times,

Fig. 11:

Generated traffic reports of the realized prototype using the example of a traffic jam event.

Fig. 12:

Generated traffic reports of the realized prototype using the example of a traffic jam event with simultaneous display of the speed measurements in the background.

• der zeitlichen Reaktion auf eingehende Daten,
• der Auflösung bei der Sicht des Systems auf die Verkehrsdomänen und
• der Häufigkeit von Meldungen ohne Beeinflussung der Domänenbildung anzupassen.

Es sind keine Streckenkenntnisse erforderlich. Es sind auch keine weiteren Modellbildungen notwendig. Falls solche Zusatzinformationen jedoch vorhanden sind, können sie zur weiteren Verbesserung der Systemergebnisse durch ergänzende Verwendung von Verkehrsmodellen leicht genutzt werden.
Die Qualität der Meldungen wird realisiert durch eine hohe örtliche und zeitliche Auflösung und einer damit hohen Aktualität, durch die Konsistenz der Meldungen, unabhängig von der Herkunft der Daten und vom jeweiligen Detektortyp, und durch eine hohe Stabilität. Es erfolgt eine frühzeitige Warnung vor staugefährdeten Streckenabschnitten durch die Meldung der Zustände dicht und stockend.
Das System basiert auf einer kontinuierlichen Beschreibung der Verkehrszustände. Sein Algorithmus kann synchrone, asynchrone und ereignisinduzierte Verkehrsdaten auch unterschiedlichen physikalischen Inhalts verarbeiten und nutzen. Die Meßstellen können ortsfest sein, es kann aber auch an variablen Orten gemessen werden.
Das System benötigt keine aufwendige Interpolation der Meßwerte entlang der betrachteten Straßen.
Die Meldungen des Systems werden zum Funktionstest mit realen historischen Verkehrsdaten und zur Überwachung im Betrieb in Kongruenz mit den Verkehrsmeßwerten visualisiert. Die Visualisierung zeigt auch die Dynamik der Verkehrslage.
Das System ermöglicht auch die Unterscheidung zwischen einem Totalstau und völlig freiem Verkehr, obwohl in diesen beiden Extremfällen jeweils keine sinnvollen Verkehrsmeßwerte vorliegen.
Nach jedem Verarbeitungszyklus liegt für jede betrachtete Straße eine Liste gemeldeter Verkehrsdomänen vor, die eine lückenlose dynamische Darstellung der Verkehrssituation des betrachteten Straßennetzes darstellt.
Grundlage eines solchen Systems sind die folgenden, auch einzeln sinnvoll verwendbaren Verfahren zur Verarbeitung und Aufbereitung von Verkehrsinformationen.The properties of an efficient traffic day classification and avoidance system according to the invention are specified in the following by way of example.
The traffic conditions stowed, stagnant, dense and free are classified by means of characteristics. If necessary, further states can be supplemented by integrating further features.
Similar traffic conditions, which were determined by local condition indicators, are combined to traffic domains. The growth, migration, division, and transitions of these domains into the other classified traffic conditions until their dissolution in the class '' free '' are dynamically followed.
Domain formation is not tied to a route-dependent or stability-related grid. The fineness of the local resolution can be selected.
The "probabilities" for the displayed traffic conditions and their significance are determined.
The message management is based on the determined dynamic of the domains and is not tied to fixed locations.
Newly determined domains can be assigned to previously reported domains using a similarity measure. Since the states of the domains are continuously described, the method also allows state transitions in the life cycle of a domain.
The algorithm offers degrees of freedom to use parameters to set the desired sensitivity of the system

• the temporal response to incoming data,
• the resolution when the system views the traffic domains and
• adapt to the frequency of reports without influencing the domain formation.

No route knowledge is required. No further modeling is necessary. However, if such additional information is available, it can easily be used to further improve the system results through the additional use of traffic models.
The quality of the messages is realized by a high spatial and temporal resolution and thus a high topicality, by the consistency of the messages, regardless of the origin of the data and the respective detector type, and by a high stability. There is an early warning of sections of the route at risk of congestion by reporting the conditions dense and stagnant.
The system is based on a continuous description of the traffic conditions. Its algorithm can process and use synchronous, asynchronous and event-induced traffic data of different physical contents. The measuring points can be stationary, but it can also be measured at variable locations.
The system does not require complex interpolation of the measured values along the streets under consideration.
The messages of the system are visualized for the function test with real historical traffic data and for monitoring during operation in congruence with the traffic measurements. The visualization also shows the dynamics of the traffic situation.
The system also enables a distinction to be made between total traffic jams and completely free traffic, although there are no sensible traffic measurements in each of these two extreme cases.
After each processing cycle, there is a list of reported traffic domains for each street under consideration, which represents a complete dynamic representation of the traffic situation of the street network under consideration.
The basis of such a system is the following, individually usable methods for processing and processing traffic information.

Traffic data preprocessing and feature creation

• v.med20 = Geschwindigkeitsmeßwerte mediangefiltert mit Fensterbreite 20 min;
• d.med20 = Dichtewerte median-gefiltert, Fensterbreite 20 min;
• f.sigma20 = Standardabweichung der Flußwerte innerhalb 20 min;
• v.tophat15 = Geschwindigkeitswerte tophat-gefiltert mit Fensterbreite 15 min.

Diese Merkmale sind Größen, die beispielsweise Meßwerte mit gemindertem Rauschanteil wiedergeben, oder die den Grad der Geschwindigkeits- und Flußschwankungen in Abhängigkeit vom Ort quantifizieren. The method according to the invention for recording the traffic situation with a history window, multidimensional morphological data filtering and feature vector formation receives traffic data from various sensors as source information and converts this via location-time filtering into features for each measurement location, which describe the local traffic situation.
Induction loops are installed stationary at different intervals on the motorway and deliver synchronously, averaged over the cycle time, the measured values for the speed, the flow of traffic, the time and the location of the measurement. Speed and flow values can also be available separately for cars and trucks.
Infrared or radar sensors are also fixed and provide event-induced asynchronous measurements.
"Floating cars" only measure the speed of one vehicle in each vehicle, i.e. at variable locations and asynchronously. The measured values are thus the speed, time and location coordinate of the measurement.
The local traffic density is calculated from these source data and the measurement vectors are formed over location x and time t. The measured values can occasionally fail, so that local and temporal gaps arise, in addition to the locations that are not covered by detectors. These data of different origins and quality are now processed into consistent and stable statements about the traffic situation.
First, incorrect data is eliminated. If the mean speed over the cycle time could not be determined due to lack of traffic, a specially defined value is displayed. However, the measured values are also missing when traffic is completely blocked. In both cases the flow is zero. A memory of the previous course of speed and flow is used to decide which situation really exists. Since very small flow and speed values can occur in small areas within a traffic jam, values below a minimum size are not used for density calculation.
All measurement vectors are displayed via sliding "history windows" for each measurement category - speed v, traffic flow f and density d - from e.g. Passed on for 20 minutes. The previous values are deleted. The location is in these history windows in small intervals of e.g. 200m discretizes the time in intervals of e.g. 1 minute. The history window is synchronized with the synchronous detector data from e.g. Continued for 1 min. In this way, the measured values are obtained as sliding time functions while maintaining the local-time relationship and with great topicality (FIG. 4).
Since the measured values are very noisy (FIG. 1), a local and temporal filtering of the speed and density values is carried out in order to reduce the fluctuations without, however, suppressing significant state transitions. The analysis of the time functions of the measurement vectors shows that a median filter is suitable for this (FIG. 2).
The standard deviation of the flow values from the history window and the dilated tophat function of the speeds with e.g. 15 minutes width used (Fig. 3) to obtain measures for the temporal fluctuation of the traffic. The larger these two characteristics are, the more likely there is traffic.
For the Tophat function, a morphological filter, the following applies (see Serra, J., "Image Analysis and Mathematical Morphology", 1982, Academic Press): Tophat (v) = v-Opening with Opening (v) = Dilatation (erosion (v)). Dilation and erosion are also morphological filters.
The filters must also work with measurement value gaps, ie in the event of missing measurement values at some location and time coordinates within the history window. With the morphological ranking operations, median, erosion and dilation, from which the tophat filtering also consists, this need is easy to take into account: nonexistent measurement values are omitted. Only existing values are also used to calculate the standard deviation of the flow values.
After this preprocessing (FIG. 4), the resulting feature vector is available for each location x: (v.med20, d.med20, f.sigma20, v.tophat15) T (X). Mean:

• v.med20 = Velocity measured values median filtered with window width 20 min;
• d.med20 = density values median-filtered, window width 20 min;
• f.sigma20 = standard deviation of the flow values within 20 min;
• v.tophat15 = speed values tophat-filtered with window width 15 min.

These features are quantities which, for example, reproduce measured values with a reduced noise component, or which quantify the degree of speed and flow fluctuations depending on the location.

This feature description is the necessary prerequisite for it high-quality processes for location- and time-resolved State classification of traffic.

• Detektordaten empfangen
  • synchron, asynchron, ereignisinduziert, ortsfest, variabler Ort
• Detektordaten verarbeiten
  • Lokale Verkehrsdichte berechnen
  • Meßvektor bilden: (v, f, d)^T (t, x)
  • Plausibilitätskontrolle, fehlerhafte Daten eliminieren
• "Historienfenster"
  • Gleitende Orts- und Zeitfunktionen der Meßwerte
• Filterung über Ort und Zeit
  • Medianfilter für Geschwindigkeits- und Dichte-Daten
  • Standardabweichung der Flußwerte
  • Dilatierte Tophat-Funktion der Geschwindigkeiten
  • Resultierender Merkmalsvektor an jedem Ort x zur Beschreibung der örtlichen Verkehrssituation: (v.med20, d.med20, f.sigma20, v.tophat15)^T (x)

The method according to the invention is therefore characterized by the following processing steps:

• Detector data received
  • synchronous, asynchronous, event-induced, stationary, variable location
• Process detector data
  • Calculate local traffic density
  • Form measurement vector: (v, f, d) ^T (t, x)
  • Plausibility check, eliminate incorrect data
• "History window"
  • Floating location and time functions of the measured values
• Filtering over time and place
  • Median filter for speed and density data
  • Standard deviation of the flow values
  • Dilated tophat function of speeds
  • Resulting feature vector at each location x to describe the local traffic situation: (v.med20, d.med20, f.sigma20, v.tophat15) ^T (x)

Classification of traffic conditions based on traffic situation descriptions

The method according to the invention for detecting the traffic status of a road traffic network by means of fuzzy classification and the formation of value-continuous location status vectors uses feature vectors which are obtained by preprocessing the measured values and describe the traffic situation at each measurement location (x). However, measured values or other calculated values, e.g. Travel times are processed. The method converts these feature vectors via a fuzzy classification into a suitable description of the traffic conditions in the form of location status vectors.
To differentiate the traffic conditions: traffic jams, traffic jams, dense and free, each traffic pattern is assigned a vector of features that represents a point in the feature space. The n-dimensional feature space is spanned by the feature axes (Fig. 5).

• Klassifikation der Verkehrszustände
  • Merkmale der Verkehrszustände: gestaut ... frei
  • Aufteilung des Merkmalraumes durch Fuzzy-Diskriminanzfunktionen
  • Fuzzy-Klassifikation der Komponenten der vorhandenen Merkmalsvektoren an jedem Ort: Wahrscheinlichkeit der Verkehrszustände je Merkmal
• Vektorielle Addition und Normierung zu einem Zustandsvektor für jeden Ort: Ortszustandvektor

The classification of the patterns now corresponds to a spatial division of the feature space. This can be done by separating functions that set exact limits. However, an approach is chosen here that uses discriminant functions for each class. A pattern is then assigned to the class whose discriminant function for the characteristics of this pattern is greater than the other discriminant functions (Duda, RO and Hart, PE: "Pattern Classification and Scene Analysis", New York 1972). Each class stands for one of the different traffic conditions.
To define the discriminant functions, a definition of the traffic conditions mentioned is necessary. Even a verbal description shows that the descriptions are subjectively colored and diffuse and that the boundaries between the traffic conditions are not clear, but rather of a qualitative nature through comparisons with the other traffic conditions. This problem, that the definition of the traffic conditions is only "unsharp", is taken into account by a fuzzy classification. The fuzzy classification functions are applied to the present feature vectors at every location (FIG. 6).
Each component is assigned a probability using the component-by-component fuzzy classification. These component state vectors are then combined by vectorial addition with subsequent normalization to form a single state vector for each location (FIG. 7).
Instead of a summation, the state probabilities can also be multiplied component by component. Instead of normalizing the vectors to the sum of the components equal to 1, normalization to the vector amount equal to 1 is also possible.
From other sources, e.g. Local state vectors determined from model-based methods are included in the same way when the feature-based state vectors are combined with the local state vectors, if necessary taking into account additional weighting factors.
At all locations where there were sufficient measured values and feature vectors could be calculated from them, there are local state vectors which contain the probabilities for the different traffic conditions. The method according to the invention for detecting the traffic state by means of fuzzy classification and location state vectors is therefore characterized by the following processing steps:

• Classification of traffic conditions
  • Characteristics of the traffic conditions: jammed ... free
  • Distribution of the feature space through fuzzy discriminant functions
  • Fuzzy classification of the components of the existing feature vectors at each location: probability of the traffic conditions per feature
• Vectorial addition and normalization to a state vector for each location: local state vector

Extrapolation to create a complete description of traffic conditions

The method according to the invention for creating a complete description of the traffic condition of a road traffic network is based on location condition vectors. The components of these standardized location status vectors represent the probabilities for the existence of a traffic status at the location in question.
In order to bridge these gaps, according to the invention an extrapolation of the locally available location state vectors is carried out by means of e.g. a local Gaussian filter over the location (convolution with a Gaussian curve or a bell curve similar to the Gaussian curve, e.g. f (x) = 1 / (1 + x ²ⁿ ), n a natural number), for each component of the local state vector Cut. The extrapolation range is determined by the parameters of the Gaussian filter. Due to this location filtering, location state vectors are now available at practically all locations (FIG. 8). However, this extrapolation may only be carried out up to a certain distance from the locally available location state vectors, so that the extrapolation values also correctly describe the real conditions. The maximum extrapolation width is dynamically limited by a threshold for the amount of extrapolated state vectors, ie not rigid, but in dependence on the extrapolation result. As the extrapolation range increases, the weighting of the existing location state vectors must also decrease. For example, the Gaussian filter.
In order to further ensure the stability of the detection of the differentiated traffic conditions, a component-wise averaging of the status vectors is also carried out at each location with the values of the previous update run. This smooth temporal smoothing based on the continuous description of the state at each location is much more effective than an artificially enforced stability over time in subsequent processing steps. Such sliding smoothing is a form of an autoregressive filter (AR) (Papoulis, A .: "Probability, Random Variables, and Stochastic Processes", McGraw-Hill Series, in: Systems Science, McGraw-Hill 1991).
In order to keep this smooth temporal smoothing approximately independent of the selected time intervals between the updates of the individual roads, the filter parameter a _{tstep is set} to a fixed interval of e.g. 10 minutes: z = a tstep z New + (1-a tstep ) z old with a tstep = 1 - (1-a t10min ) tstep / t10min

• Extrapolation jeder Komponente der Ortszustandsvektoren über den Ort durch Filterung mit z.Bsp. einem lokalen Gaußfilter.
• Gleitende (autoregressive) Mittelwertbildung der Ortszustandvektoren über die Zeit.

Z _new or z _old stand for the newly determined or in a previous calculation cycle determined state vector components of a street, and tstep and t10min stand for the selected time interval in seconds between 2 update runs of the street concerned and for 600 seconds respectively. The z calculated in this way then becomes _old for this road in the next update run.
So there are two processing steps:

• Extrapolation of each component of the location state vectors over the location by filtering with e.g. a local Gaussian filter.
• Moving (autoregressive) averaging of the position state vectors over time.

Comparison of traffic condition descriptions by probabilistic similarity measures

The method according to the invention for comparing traffic condition descriptions and traffic domains by means of probabilistic similarity measures is based on location condition vectors which, for example, from feature vectors for each measuring location via a fuzzy classification and a subsequent weighted extrapolation for each selectable discretization of the distance between the measuring points, e.g. every 200m. The components of these location state vectors are the probabilities for the presence of the differentiated traffic conditions at every location. The location state vectors are thus a continuous description of the traffic conditions at each location under consideration and that for each location of the respective route section and thus for the entire road network.
Places with similar traffic conditions, that is, similar location status vectors, are combined into traffic domains in order to be able to represent the traffic situation of the entire road traffic network. For each of these traffic domains, their state vector results from vectorial addition and subsequent normalization of the location state vectors of all locations which are combined in this domain.
The definitive determination of the traffic condition at the location under consideration, the combination of the locations of the same traffic condition into domains and the evaluation of the domains among each other presupposes that an objective comparison of the location status vectors with each other and the domain status vectors with each other is possible. According to the invention, probabilistic similarity measures are used for this.
The comparison of the location state vectors is carried out component by component on the basis of the probabilities contained in the vector components for the presence of the differentiated traffic conditions at the location under consideration.
The basis for comparison is a similarity measure, for which a metric, i.e. a distance measure, is used. A similarity measure is also used according to the invention for comparing the domains with one another. The similarity measure for the traffic domains is also a metric that uses a state vector metric for the domain state vector and also takes into account the local position and length of the domain under consideration.
The state of maximum probability, in particular the index of the maximum component of a state vector relative to the other components, can be used as a measure of similarity for both state vectors, the location state vector or the domain state vector.
The Euclidean vector norm can also be used as a metric for the state vectors: Sum ((vek1- vek2) ² ).
The location of the traffic domains, which is included in the metric as a measure of similarity, can be taken into account as follows:
Amount (middle.domain1 - middle.domain2) / (length.domain1 + length.domain2) * 2. The following can be used as a metric for the length of the traffic domain:
Amount (length.domain1 - length.domain2) / max (length.domain1, length.domain2).

Dynamic domain formation of the same traffic conditions

The method according to the invention for determining domains of the same traffic conditions is based on location state vectors, which for example. from feature vectors for each measuring location via a fuzzy classification and a subsequent weighted extrapolation for each selectable discretization of the distance between the measuring points, e.g. every 200m. The components of these location state vectors are the probabilities for the presence of the differentiated traffic conditions at every location. The location state vectors are thus a continuous description of the traffic conditions at each location under consideration and that for each location of the respective route section and thus for the entire road network (FIG. 9 left).
Places with similar traffic conditions, that is, similar location status vectors, are combined into traffic domains in order to be able to represent the traffic situation of the entire road traffic network. The comparison of the traffic condition descriptions by the location condition vectors is carried out component by component by means of a probabilistic similarity measure, for which a metric, that is to say a distance measure, is used.
For each of these traffic domains, their domain state vector results from vectorial addition and subsequent normalization of the location state vectors of all locations that are combined in this domain.
The domains are delimited from one another in such a way that stable domain tracking is possible. This segmentation problem is solved as follows:
For each location, the traffic state with the maximum probability is determined from its state vector.
Starting with the location whose state vector e.g. "Congestion" indicates in both directions whether the components of the state vectors of the neighboring locations are most likely to show the same traffic state or the component with the traffic state to be compared is at most below the maximum component by the hysteresis value. Due to this hysteresis, places whose state vectors speak for themselves another traffic state, but differ only slightly from the state of the starting point, are also counted as "traffic jam domain". The state vector of the traffic domain found in this way is the normalized sum of the state vectors of the locations of the domain. For example, the sum of the components equal to 1 is selected as the norm. The focus of the domain is the sum of the location coordinates of the domain, which are weighted with the amount of the corresponding local location state vectors.

• Domänenbildung
  • Zusammenfassung der Orte mit ähnlichen Zustandsvektoren mit "Hysterese" zu Domänen
  • Domänenzustandsvektoren: vektorielle Addition und Normierung der Zustandsvektoren der Orte der Domänen
  • Ermittlung: Domänenschwerpunkt, Domänenlänge
• Resultat: Domäneneinteilung des Straßenverkehrsnetzes.
• Permanente Aktualisierung in zeitlichen Intervallen

In the same way, the domains for the classes of traffic conditions are stagnant, dense and free.
Alternatively, the domain formation can also be found by means of a cluster analysis with a subsequent relaxation process over the location. The distances between the state vectors and their spatial distance can serve as a similarity measure for the cluster search. The result of this method is local agglomerations with similar traffic conditions within the agglomerations. However, this procedure is much more computationally complex.
The domain list may contain neighboring domains that are very similar in terms of their domain state vector due to the method of their formation. These are summarized. The list of domains is sorted for each street according to the geographic focus of the domains. Domains are summarized if they are separated by a gap up to a maximum length and their state probabilities are the same or differ at most by the specified hysteresis amount as shown above.
The result is a current list of traffic domains for a street or the road network. The entire road network is broken down into the routes of the domains. The domains contain, among other things, a continuous status description. The domain division of the road network is repeatedly determined and updated at time intervals (FIG. 9, right).
The method according to the invention for dynamic domain formation of the same traffic conditions is therefore characterized by the following processing steps:

• domain formation
  • Summary of locations with similar state vectors with "hysteresis" to domains
  • Domain state vectors: vectorial addition and normalization of the state vectors of the locations of the domains
  • Determination: domain focus, domain length
• Result: Domain division of the road network.
• Permanent update at intervals

Dynamic domain tracking and message management

The method according to the invention for the dynamic tracking of domains of the same traffic conditions is based on location condition vectors which consist of the feature vectors for each measurement location and by weighted extrapolation for each selectable discretization of the route between the measurement points, e.g. every 200m. The components of these location state vectors are the probabilities for the presence of the differentiated traffic conditions at every location. The location state vectors are thus a continuous description of the traffic conditions at each location under consideration and that for each location of the respective route section and thus for the entire road network.
Places with similar traffic conditions, that is, similar location status vectors, are combined into traffic domains in order to be able to represent the traffic situation of the entire road traffic network. The comparison of the traffic condition descriptions by the location condition vectors is carried out component by component using a probabilistic similarity measure, for which a metric, that is to say a distance measure, is used.
For each of these traffic domains, their domain state vector results from vectorial addition of the location state vectors of all locations which are combined in this domain.
The result is a current list of traffic domains for a street or the road network. The entire road network is broken down into the routes of the domains. The domain state vectors contain, among other things, a continuous one. Grading. The domain division of the road network is repeatedly determined and updated at time intervals.
The method according to the invention for dynamic tracking of these domains works as follows:
The current domains determined in an update run are compared with the domains found in the previous update run, each of which has been stored in a message list. The newly found domains are assigned to the existing domains using a similarity measure. The basis of the similarity measure is the continuous domain state vectors as well as the position and length of the stretches that are taken up by the domains. If a minimum similarity is found, the domain route under consideration is assigned to the corresponding message or the already existing domain. This is done for all newly discovered domains and existing messages.

This assignment based on a continuous similarity measure for the continuous domain state vectors also allows a gradual change in the traffic state over the course of a message.
Each of these assignments between domains and messages is checked. Which previous messages remain in the message list, whether there are change notices because the location and / or the length of the previous domains have changed due to the assignment of newly found domains, and which unassignable domains have to be reported again. A new message may be generated for new domains, the message attributes for domains assigned to existing messages are updated and change messages may be generated, and messages which could not be assigned to a domain or only to a domain with the traffic state "free" are deleted.
A hysteresis parameter can be used to control when new and change reports are considered significant and actually executed. This enables the number of messages to be reduced without restricting the internal information for domain formation and tracking of the process.
The result of the method according to the invention is a dynamic tracking of the domains with similar traffic conditions with each update step and a constantly updated message list.
The list of messages with the domain properties contained in each message is the basis for information about the traffic conditions on the road network. This information is stored until the next update step and is brought to the attention of the vehicle drivers in a suitable manner, e.g. via mobile radio or signal systems. or they are used for control interventions in traffic or the direct influencing of vehicles. The information that is passed on to the road users can also be given individually depending on their respective positions and destinations in the road network.

visualization

For the functional test of the methods with real historical traffic data and for the optical monitoring of the methods during operation as well as for the visual representation of the traffic conditions on road sections, the message results of the methods are visualized according to the invention and compared with the measured traffic values. For this purpose, the measured traffic values in the section of road under consideration are represented by different gray values over time. The reports of the traffic conditions or the domains are also shown in this coordinate system. With the time axis moving over the image, one can recognize the traffic situation and its dynamics and from the congruence between measurement and message the quality of the process results (FIG. 10).
Examples of this; which creates a realized prototype are shown in FIGS. 11 and 12.

Over the entire location, the algorithm is executed repeatedly over time. In the figures, the update section therefore moves from left to right. To do this, the saved values to the left of the current time of evaluation are viewed. The locations of domains found are shown by vertical lines. If messages are activated for these domains, this is indicated by a white spot in the focus of the domain. A connecting line is then also drawn between the focal points of the domain, which visualize the temporal assignment by the method for dynamic tracking of the domains. The different gray values of the lines correspond to different colors with which the states are reproduced, stagnant, dense and freely reproduced.
Small diamonds can also be seen, which show the local traffic conditions at the locations where traffic measurements are available.
The traffic measurements are shown in gray in the background. For example, in the figures. to see the speed values, with dark gray values representing low speeds, and correspondingly light gray values representing high speeds. You can immediately see how the messages generated are in relation to the traffic measurements in place and time.
In the traffic situation, which is shown in FIG. 11, the domain is reported as dense at the bottom left, for example. This message changes to jammed shortly thereafter (bottom transverse line). These domains can be tracked stably over a longer range in x and t until the end of the period under consideration.
Above, with larger x values, a domain of stagnant traffic is recognized somewhat later (middle line), which can also be followed very stably over time.
Even later, a smaller domain of dense traffic is recognized.

• Dynamische Verfolgung der Domänen
  • Bei jedem Aktualisierungsschritt:
  • Zuordnung der neuen Domänen über ein Ähnlichkeitsmaß zu den vorhandenen Meldungen
  • Resultat: Liste der Domänen und Meldungsliste auf der jeweiligen Straße
• Meldungsmanagement
  • Meldungsliste: Informationen über Verkehrszustände und Verkehrslage
  • Aktualisierung der Meldungsliste bei jedem Aktualisierungslauf Falls Zuordnung neuer Domänen in der Meldungsliste möglich:
    • bisherige Meldung bleibt bestehen oder Änderungsmeldung bezüglich Lage, Länge und Verkehrszustand
  • Falls Zuordnung in der Meldungsliste nicht möglich:
    • Neumeldung oder Löschung
• Visualisierung
  • Optische Darstellung der Kongruenz zwischen Verkehrsmeßwerten und Meldungungen
  • Dynamik der Domänen bzw. der Verkehrszustände
• Überblick über die gesamte Verkehrslage

In FIG. 12, the messages generated are shown together with the measured speed values in the background.
This type of representation gives a visual impression of the traffic dynamics and an overview of the entire traffic situation of the road network.
A prototype of the visualization process was realized and tested in practical operation.
The method according to the invention for the dynamic tracking and assignment of domains of the same traffic conditions and the message management are therefore characterized by the following processing steps:

• Dynamic domain tracking
  • At each update step:
  • Assignment of the new domains via a similarity measure to the existing messages
  • Result: List of domains and message list on the respective street
• message management
  • Message list: Information about traffic conditions and traffic conditions
  • Update of the message list with every update run If new domains can be assigned in the message list:
    • the previous message remains or a change message regarding location, length and traffic condition
  • If assignment in the message list is not possible:
    • New notification or deletion
• visualization
  • Optical representation of the congruence between traffic measurements and messages
  • Dynamics of domains and traffic conditions
• Overview of the entire traffic situation

Essential details

• Die Verarbeitung der Verkehrsmeßdaten von unterschiedlichen Sensoren, die ortsfest und mobil sind, synchrone, asynchrone und ereignisinduzierte Meßwerte liefern, lückenhaft und sehr verrauscht sind.
• Die Klassifizierung der verschiedenen Verkehrszustände: gestaut, stockend, dicht und frei, die nicht eindeutig definierbar sind, und die dynamische Bildung und Verfolgung von Domänen gleicher Verkehrszustände, mit hoher örtlicher und zeitlicher Auflösung, auch ohne Streckenkenntnisse und ohne Bindung an ein systembedingtes, streckenabhängiges Raster.
• Ein Meldungsmanagement, das stabil und konsistent die Verkehrszustände bzw. Verkehrsdomänen anzeigt, permanent aktualisiert wird und auch das Wachstum der Domänen, ihre Wanderung und Teilung sowie ihre Übergänge in die anderen klassifizierten Verkehrszustände bis zu ihrer Auflösung in der Klasse "frei" darstellt.

Die Betrachtung des Standes der Technik und die kritische Analyse der bekannten Verfahren zeigen, daß es für diese Problemstellung noch keine befriedigende Lösung gibt.
Das erfindungsgemäßen Verfahren lösen diese Aufgabenstellungen, die durch die folgenden wesentlichen Verarbeitungsschritte charakterisiert sind:

• Aus den Quelldaten der Sensoren wird die lokale Verkehrsdichte ermittelt und für jeden Meßort und Meßzeitpunkt ein Meßvektor gebildet.
• Die Komponenten der Meßvektoren werden über mit den Aktualisierungsschritten gleitende Historienfenster als gleitende Orts-Zeitfunktionen dargestellt.
• Es folgt eine komponentenweise orts-zeitliche Filterung der Meßvektoren, aufgrund der Analyse der Zeitfunktionen hier bevorzugt mittels Medianfilter, Bildung der Standardabweichung und der dilatierten Tophat-Funktion. Es ergibt sich an jedem Ort ein resultierender Merkmalsvektor, der die lokalen Verkehrssituationen beschreibt.
• Den orts-zeitlich gefilterten Meßgrößen werden nun über eine Fuzzy-Klassifikation Wahrscheinlichkeiten für die Verkehrszustände zugeordnet. Damit ist es möglich, diese Komponenten der Zustände zu summieren und normiert zu einem einzigen Zustandsvektor für jeden Ort konsistent zusammenzuführen.
• Sind an einem Ort zu wenig Meßwerte vorhanden, gibt es dort auch keinen Zustandsvektor, es sind Lücken vorhanden. Zur Überbrückung wird eine Extrapolation mit z.Bsp. einem lokalen Gaußfilter vorgenommen. Die Stabilität wird weiter erhöht durch eine gleitende, komponentenweise Mittelung der Zustandsvektoren von jeweils zwei zeitlichen Aktualisierungsschritten (Autoregressives-Filter).
• Orte mit ähnlichen Zustandsvektoren werden zu Domänen zusammengefaßt. Durch vektorielle Addition der lokalen Zustandsvektoren und anschließender Normierung wird der Domänenzustandsvektor ermittelt. Es resultiert eine aktuelle Liste von Domänen mit kontinuierlicher Verkehrszustandsinformation für jede betrachtete Straße.
• Bei jedem weiteren Aktualisierungsschritt werden die identifizierten Domänen den bereits vorhandenen Domänen der Meldungsliste über ein kontinuierliches Ähnlichkeitsmaß zugeordnet. Die Domänen werden also über die Zeit dynamisch verfolgt, wobei auch eine graduelle Zustandsänderung über die Zeit möglich und erlaubt ist.
• Nach dieser Zuordnung wird die Meldungsliste für die betrachtete Straße aktualisiert, und es werden je nach Signifikanz Neu-, Änderungs- und Löschmeldungen ausgeführt. Die Signifikanz wird über einen Parameter gesteuert, womit die Meldungszahl ohne Beeinträchtigung des Verfahrens reduziert werden kann. Es resultiert eine aktuelle Meldungsliste je Straße.
• Die Meldungslisten sind die Basis für das Meldungsmanagement. In ihnen werden die Informationen über die Verkehrszustände gespeichert und stehen für die genannten Anwendungen zur Verfügung.
• Die Ergebnisse werden visualisiert. Die Darstellung kann der Überwachung dienen und gibt einen optischen Eindruck über die Verkehrsdynamik und einen Überblick über die gesamte Verkehrslage.

Die Verfahren wurden erprobt.The invention relates to methods which can be used independently of one another, but which can also be combined to form a traffic situation classification and reporting system for road traffic, preferably for motorways and expressways.
There is an urgent need for timely and up-to-date information about the differentiated traffic conditions in sections of the route that are identified as precisely as possible for safety reasons, but also for numerous other services and, if necessary, for direct vehicle control and traffic control. Key problems that need to be solved are:

• The processing of traffic measurement data from various sensors that are stationary and mobile, deliver synchronous, asynchronous and event-induced measured values, are incomplete and very noisy.
• The classification of the different traffic conditions: congested, stagnant, dense and free, which cannot be clearly defined, and the dynamic formation and tracking of domains of the same traffic conditions, with high spatial and temporal resolution, even without route knowledge and without being linked to a system-related, route-dependent grid ,
• A message management that shows the traffic conditions or traffic domains in a stable and consistent manner, is constantly updated and also shows the growth of the domains, their migration and division as well as their transitions into the other classified traffic conditions until they are resolved in the "free" class.

The examination of the prior art and the critical analysis of the known methods show that there is still no satisfactory solution to this problem.
The method according to the invention solves these problems, which are characterized by the following essential processing steps:

• The local traffic density is determined from the source data of the sensors and a measurement vector is formed for each measurement location and measurement time.
• The components of the measurement vectors are represented as history-time functions which slide with the update steps.
• A component-by-location filtering of the measurement vectors follows, based on the analysis of the time functions here preferably by means of a median filter, formation of the standard deviation and the dilated tophat function. There is a resulting feature vector at each location, which describes the local traffic situations.
• Probabilities for the traffic conditions are now assigned to the spatially-temporally filtered measured variables via a fuzzy classification. This makes it possible to sum these components of the states and to consistently merge them into a single state vector for each location.
• If there are too few measured values at one location, there is no state vector there, there are gaps. An extrapolation with e.g. a local Gaussian filter. The stability is further increased by a sliding, component-wise averaging of the state vectors of two temporal update steps (autoregressive filter).
• Locations with similar state vectors are combined into domains. The domain state vector is determined by vectorial addition of the local state vectors and subsequent normalization. The result is a current list of domains with continuous traffic status information for each street under consideration.
• With each further update step, the identified domains are assigned to the existing domains of the message list using a continuous measure of similarity. The domains are thus tracked dynamically over time, and a gradual change in state over time is also possible and permitted.
• After this assignment, the message list for the street under consideration is updated and new, change and deletion messages are carried out depending on the significance. The significance is controlled via a parameter, with which the number of messages can be reduced without affecting the procedure. The result is a current message list for each street.
• The message lists are the basis for message management. The information about the traffic conditions is stored in them and is available for the applications mentioned.
• The results are visualized. The display can be used for monitoring and gives a visual impression of the traffic dynamics and an overview of the entire traffic situation.

The procedures have been tried.

Claims (48)

1. Method of producing traffic information representing the traffic situation in a road traffic network, in which the traffic information is processed by measured traffic values detected at several different times, and the detected measured traffic values for each road observed are entered and stored - in terms of the place (x) and time (t) of their detection for each measured value category - in repeatedly updated history windows,
   characterised

   a) in that the history windows extend for a specific period into the past from the current time of the production of the traffic information, and discretise the time and place at intervals, whereby the detected measured traffic values in the individual history windows which are currently being viewed are filtered, using a variety of filters, in terms of their time and location progressions, a characteristic being formed for each filter from which, for a single location in the road network in each case, there is produced - by combining the individual characteristics - a location status vector which relates to this location and describes the traffic situation comprehensively,

   b) in that the measured traffic values of the traffic situation of the road traffic network are recorded by fixed sensors positioned along roads in the road network, and/or measured by mobile sensors within vehicles travelling on the road traffic network,

   c) in that, in the history windows for the individual measurement categories, one- and two-dimensional morphological filters for place and time are applied to the measured traffic values relating to place and/or time which are present in incomplete form,

   d) in that it produces traffic information which is capable of being issued, based on the location status vectors which describe the local traffic situation,

   e) furthermore in that locations with similar location status vectors, and thus similar traffic statuses, are combined into traffic domains with freely definable ends,

   f) in that an allocation to existing domains is carried out on the domains arising from each further update stage, by means of a continuous similarity measure, for the purpose of dynamic tracking - in particular in transitional situations - and so that a possible gradual change in status can take place over time,

   g) in that the domain partitioning of the road traffic network is specified repeatedly at intervals of time,

   h) in that traffic reports are produced on the basis of the domain partitioning

   i) and in that the traffic information is issued, by means of a transmitter or by access to a transmitter for broadcasting traffic information, in particular by mobile phone, to road users or to a service provider.
2. Method as in Claim 1
   for detecting the traffic status of a road traffic network by forming location status vectors from the characteristics vectors that describe the traffic situation at each measuring location, or directly from measured traffic values at the measuring locations,

   whereby, for each location, the characteristics that describe the traffic situation at this location are evaluated by means of a fuzzy discrimination function per characteristic for each possible classifiable traffic status (Fig. 6: jammed, stop-start, heavy, free-moving), the result being a value-continuous status vector for each location and each characteristic (Fig. 7, right), the components of which quantify the probability of the presence of a traffic status,

   after which, existing status vectors for each location are combined into a single status vector for each location, and the resulting status vector for each location is standardised as the location status vector (Fig. 7),

   whereby traffic information which is capable of being issued is produced on the basis of the location status vectors.
3. Method as in one of the preceding Claims
   for producing a comprehensive traffic status description representing the traffic situation in a road traffic network, in the form of location status vectors, the components of which quantify the probability of the presence of a traffic status,

   whereby gaps in the location status descriptions of locations (caused by the absence of measured traffic values at these locations) are covered by extrapolating the existing location status vectors by means of a filtering process in the form of a convolution over the existing location status vectors (Fig. 8),

   whereby traffic information which is capable of being issued can be produced by means of traffic status vectors from the now complete description of traffic conditions across the location.
4. Method as in Claim 1
   for comparing traffic information in the form of status vectors, the components of which quantify the probability of the presence of a traffic status,

   whereby the status vectors describe the traffic situation of a road traffic network by means of similarity measures which represent the similarity of traffic statuses,

   whereby metrics are used as the similarity measure between two status vectors.
5. Method as in one of the preceding Claims 1-4,
   for the constant dynamic tracking of domains characterised by the same traffic statuses within a domain, and the allocation of domains to domains, already identified by earlier traffic messages, in a road traffic network,

   whereby the division into domains of the road network is established repeatedly at intervals of time, the domains found at the current time for each road under observation being compared, by means of a similarity measure which continuously evaluates the section and status information belonging to the domains, with the domains established in the previous evaluation cycle of the roads concerned (Fig. 10),

   after which new messages are sent, using a message management system, for domains which have been newly added compared with the previous evaluation cycle, cancel messages are sent for domains which have disappeared, and

   change messages are sent for all domains which it has been possible to allocate to a section which has already been reported, the messages being sent to the road users or telematics service providers, section information produced on this basis - consisting of situation, status and length of the traffic domains of an entire observed road in each case - being used to generate the traffic information which is issued in the new, change and cancel messages.
6. Method as in one of the preceding Claims 1-5,
   characterised in that
   a domain status vector, representing the traffic situation in these domains, is calculated from the location status vectors of the combined locations and is assigned to a domain.
7. Method as in one of the preceding Claims 1-6,
   characterised in that
   the average value of the location status vectors of the locations combined in this domain is used to calculate the relevant domain status vector.
8. Method as in one of the preceding Claims 1-7,
   characterised in that
   the traffic information is produced for direct transmission to road users in the form of new, change and cancel messages.
9. Method as in one of the preceding Claims 1-8,
   characterised in that
   the division into domains of the road network is established repeatedly at intervals of time.
10. Method as in one of the preceding Claims,
    characterised in that
    the traffic information is produced - in the form of new, change and cancel messages

    in a form which is suitable, without any further processing, to indirectly influence traffic and/or to directly manage the traffic.
11. Method as in one of the preceding Claims,
    characterised in that
    the traffic information is used to produce messages to road users in the road network relating to the current traffic status and/or a predicted traffic status and/or instructions for navigation, taking into account the current and/or predicted traffic status and/or speed instructions.
12. Method as in one of the preceding Claims,
    characterised in that
    the measured traffic values from at least some of the sensors are measured, chronologically synchronised with each other, at regular intervals of time.
13. Method as in one of the preceding Claims,
    characterised in that
    measured traffic values, at least from some of the sensors, are induced by events and are available or measured only when predetermined events occur.
14. Method as in one of the preceding Claims,
    characterised in that
    chronologically asynchronous measured traffic values are present from at least some of the sensors.
15. Method as in one of the preceding Claims,
    characterised in that
    characteristics from traffic data with different physical contents (flow f, density d, speed v in Fig. 4) are formed and brought together as a classification event consistently over value-continuous probabilities.
16. Method as in one of the preceding Claims,
    characterised in that
    the speed characteristic which is to be classified is calculated by derived values, in particular by travel speeds.
17. Method as in one of the preceding Claims,
    characterised in that,
    as a filtering function for calculating a characteristic for stop-start traffic, a morphological top-hat filtering of measured traffic values is used in the speed measured value category.
18. Method as in one of the preceding Claims,
    characterised in that,
    as a filtering function for smoothing out the measured traffic values in the speed and density measured value categories, a morphological median filtering process is used.
19. Method as in one of the preceding Claims,
    characterised in that
    local resolution can be selected at will for the local discretising of the history windows and subsequent processing.
20. Method as in one of the preceding Claims,
    characterised in that
    status vectors of all characteristics for each location are combined to form the location status vector for each location (Fig. 7).
21. Method as in one of the preceding Claims,
    characterised in that
    status vectors available also from other sources for each location are included in the process of combination into location status vectors, taking weighting factors into account.
22. Method as in one of the preceding Claims,
    characterised in that

    the combining of the characteristics-based status vectors for each location is effected by adding these vectors for each location, and then a standardisation on to total = 1 is carried out, or

    the combining of the characteristics-based status vectors for each location is effected by a component-wise product of these vectors for each location, and then a standardisation on to value = 1 is carried out.
23. Method as in one of the preceding Claims,
    characterised in that
    the fuzzy discrimination function is applied to locations which are independent of the fixed sectional division of the road traffic network segments of a digital map.
24. Method as in one of the preceding Claims,
    characterised in that,
    above a threshold for the amount of the extrapolated status vectors, the maximum extrapolation breadth is restricted.
25. Method as in one of the preceding Claims,
    characterised in that
    the filtering process is carried out by means of a Gaussian filter (convolution with a Gaussian curve).
26. Method as in one of the preceding Claims,
    characterised in that,
    for each road under observation, a sliding chronological smoothing of the currently established status vectors for each location is carried out with the extrapolated location status vectors established for each location for that road at the previous updating stage.
27. Method as in one of the preceding Claims,
    characterised in that
    an autoregressive (AR) filter is used for this smoothing process.
28. Method as in one of the preceding Claims,
    characterised in that
    the filter parameters of the AR filter are converted in dependence on the length of the update intervals in such a way that the filtering process is independent of the update intervals selected.
29. Method as in one of the preceding Claims,
    characterised in that
    the status of maximum probability is used as the similarity measure for the location status vectors, in particular the index of the maximum components of a status vector relative to the other components.
30. Method as in one of the preceding Claims,
    characterised in that
    the euclidian vector norm is used as the metrics for the location status vectors.
31. Method as in one of the preceding Claims,
    characterised in that
    the metrics used as the similarity measure for traffic domains apply metrics for status vectors to the domain status vectors, and also take into account at least the local situation and/or the length of the domains.
32. Method as in one of the preceding Claims,
    characterised in that
    the domain growth process is applied directly to the value-continuous extrapolated status vectors of the locations of a road which is being observed.
33. Method as in one of the preceding Claims,
    characterised in that
    the traffic situation is indicated by a traffic domain which contains the position in the section network, the length, and the traffic status in - in particular - four stages (jammed, stop-start, heavy, free-moving) as probabilities in the form of the domain status vector.
34. Method as in one of the preceding Claims,
    characterised in that
    where there are locations whose status vector in itself indicates a different status from that of the domains observed for the purpose of domain formation, the domain formation is able to allocate these locations to these domains.
35. Method as in one of the preceding Claims,
    characterised in that
    the components of the location status vectors are interpreted as continuous probability values instead of discrete values (0/1) for traffic statuses prevailing at these locations.
36. Method as in one of the preceding Claims,
    characterised in that,
    by means of a hysteresis value - relating to the status vector components - which is to be added, the level of combination of locations is controlled with a different maximum status component from that which is currently being considered for domain formation.
37. Method as in one of the preceding Claims,
    characterised in that,
    by referring to earlier measured traffic values, a distinction is made, using a local memory for the measured traffic values already received, between "no traffic" and "complete blockage".
38. Method as in one of the preceding Claims,
    characterised in that,
    for the allocation of newly established domain information to existing domain information, a similarity measure is used which continuously takes into account the length, location and status vector of these domains.
39. Method as in one of the preceding Claims,
    characterised in that
    the sensitivity of the method with regard to the consideration of rapid changes in traffic events when producing traffic information in the form of characteristics vectors can be adjusted by means of the filter parameters of the measured value filters and of the autoregressive filter.
40. Method as in one of the preceding Claims,
    characterised in that
    the sensitivity of the method with regard to the local resolution of the extrapolated location status description can be adjusted by means of the filter parameters of the extrapolation filter.
41. Method as in one of the preceding Claims,
    characterised in that
    the sensitivity of the method with regard to the tendency to combine domains, and thus to reduce the number of messages, can be adjusted by means of the filter parameters of the extrapolation filter.
42. Method as in one of the preceding Claims,
    characterised in that
    status transitions during the life of a message are permitted.
43. Method as in one of the preceding Claims,
    characterised in that
    the number of messages generated can be reduced if necessary by means of the significance criteria of changes relating to position, length and traffic status of the message information.
44. Method as in one of the preceding Claims,
    characterised in that
    the issue of traffic information is effected on an event-induced basis in the case of certain predetermined events.
45. Method as in one of the preceding Claims,
    characterised in that
    a visualisation is provided of the traffic situation and of its dynamics, and the message results are provided in congruence with the measured data for the purpose of monitoring operations in a place-time coordinate system (Fig. 10, Fig. 11, Fig. 12).
46. Central unit for detecting traffic situations, in particular for implementing the method as in one of the preceding Claims, having

    an input device for entering measured traffic values into the traffic situation detection centre,

    a memory for measured traffic values entered,

    an output device for issuing traffic information to road users,

    a device for processing measured traffic data, designed in such a way that it operates in accordance with one of Claims 1 to 45.
47. Central unit for detecting traffic situations as in Claim 46,
    characterised in that
    the input device is a receiver, or access to a receiver, for receiving measured traffic values transmitted by the sensors, in particular by means of mobile telephony.
48. Central unit for detecting traffic situations as in Claim 46 or 47,
    characterised in that
    the output device is a transmitter, or access to a transmitter, for transmitting traffic information, in particular by means of mobile telephony, to road users or to a service provider.

Images