WO2011057323A1

WO2011057323A1 - Method and system to aid craft movement prediction

Info

Publication number: WO2011057323A1
Application number: PCT/AU2010/001444
Authority: WO
Inventors: Brenton Cooper; Viru Gajanayake; David Blockow
Original assignee: Bae Systems Australia Limited
Priority date: 2009-11-10
Filing date: 2010-10-29
Publication date: 2011-05-19
Also published as: EP2499625A1; AU2010317648B2; EP2499625A4; AU2010317648A1

Abstract

Embodiments of the present invention provided a method and system to aid craft movement prediction. A path monitor obtains data position data samples for at least one craft. A pattern matcher in data communication with an exemplar path repository can identify, based on one or more craft position data samples, any one or more candidate exemplar paths from the repository of exemplar paths, and determine a path match likelihood for each candidate exemplar path. Each exemplar path is representative of a previously characterised craft movement path.

Description

METHOD AND SYSTEM TO AID CRAFT MOVEMENT PREDICTION Technical field

The technical field of the invention is external tracking of moving objects, such as aircraft and other vessels, and path prediction for the objects based on the tracked movements.

Background

A recognised air picture (RAP) is used to identify all aircraft within a given air space. Typically aircraft are identified as hostile or friendly and preferably other information such as aircraft type, size, flight number and flight plan is also included in the RAP. The RAP is typically compiled using data from a number of different sources which can include military and civilian radar, air traffic controllers, and air command and control systems. The RAP is important vital for situational awareness for pilots and flight controllers for commercial, general and military aviation. Situational awareness enabled by an accurate RAP is vital in particular for military aviation and battle control.

Known recognised air picture production systems use data such as flight numbers and flight plans supplied by military and civilian air traffic controllers for identification of aircraft detected in the given air space. However, some aircraft detected in the airspace may not be readily identifiable based on air traffic control data. Identification of such aircraft and classification of unidentified aircraft as hostile or friendly is currently performed by highly skilled human operators. The individual operator's skill and knowledge is applied to whatever data may be available in order to identify and classify unidentified aircraft. This can cause problems with operator overload and errors particularly in situations which demand operators make critical decisions or many decisions quickly, for example during battle or emergency situations. There is a need for improved systems for assisting operator decision making.

Summary

According to a first aspect to the invention there is provided a method of predicting a craft movement path comprising the steps of:

obtaining one or more craft position data samples for the craft;

identifying, based on one or more of the craft position data samples, any one or more candidate exemplar paths from a repository of exemplar paths each defined using a sequence of spatial data elements representative of a previously characterised craft movement path; and

determining a path match likelihood for each candidate exemplar path.

In an embodiment the step of identifying candidate exemplar paths comprises the steps of:

selecting a first craft position from the craft position data;

searching the repository of exemplar paths based on the first craft position to identify any exemplar paths proximate the first craft position; and

for each identified proximate exemplar path:

determining relative heading and relative path distance between the exemplar path and craft movement path; and

assessing whether or not to select the exemplar path as a candidate path based on candidate criteria.

In an embodiment the step of determining relative heading between the exemplar path and craft movement path comprises:

selecting at least one further craft position from the craft position data;

determining the distance between each further craft position and a position of the exemplar path which corresponds to the further craft position sample; and

determining the relative heading based on difference in distance between exemplar path and craft movement path at the first vessel position and further vessel position.

In an alternative embodiment the step of determining relative heading between the exemplar path and craft movement path comprises determining heading difference between an absolute heading for the craft movement path obtained from the craft movement path data at the first position and an absolute heading for the exemplar path at a corresponding position.

Exemplar paths which pass within a given distance of the first craft position can be identified as proximate the first craft position. The given distance can vary between exemplar paths.

In some embodiments the given distance is mathematically defined for each exemplar path. For example, the given distance can be a function of a standard deviation for the exemplar path. The standard deviation can be determined from known craft movements used for characterisation of the exemplar path.

Examples of candidate criteria can be applied for selection of a candidate exemplar path include but are not limited to one or more of relative heading and relative path distance between the exemplar path and craft movement path. In some embodiments an exemplar path is excluded where the heading difference between the exemplar path and craft path is greater than a given threshold angle.

In some embodiments the number of candidate exemplar paths is limited to a given number n. In such embodiments the n candidate exemplar paths having closest compliance with candidate criteria can be selected.

In some embodiments the method is performed using real time captured craft position data for a detected craft.

The step of determining a path match likelihood for each candidate exemplar path can include the steps of calculating a probability of match between the craft movement path and each candidate exemplar path, and ranking the candidate exemplar paths based on relative probability.

The path match likelihood for a given number of exemplar paths can be presented to an operator for assisting the operator to characterise or identify the nature of the craft based on the craft's movement path.

Some embodiments of the method further comprise the steps of periodically obtaining further craft position data, comparing with a chosen exemplar path and alerting an operator where deviation from the chosen exemplar path is identified.

An embodiment further comprises the step of providing the repository of exemplar paths. Providing the repository of exemplar paths in some embodiments comprises the steps of:

analysing historical data representing a plurality of tacked craft paths;

identifying common sequences of movement between regions;

identifying clusters of movement paths for the sequence; and

for each identified cluster:

analysing the cluster to determine a characterising exemplar path comprising a characterising craft movement path and craft distribution about the characterising craft movement path for the cluster; and

storing the exemplar path in the repository.

In some embodiments analysing a cluster to characterise an exemplar path comprises the steps of:

dividing each movement path of the cluster into a sequence of equidistant segments positions along the movement path;

determining a characterising craft position and distribution for each segment based on the distance between each movement path of the cluster for the segment.

The characterising craft position and distribution can be determined based on root mean square analysis of the distance between each movement path. In some embodiments the exemplar path repository is a database storing exemplar path data searchable using geospatial database queries.

According to another aspect of the invention there is provided a craft movement prediction aid system comprising:

a path monitor adapted to obtain for at least one craft one or more craft position data samples; and

a pattern matcher in data communication with an exemplar path repository and adapted to identify, based on one or more craft position data samples, any one or more candidate exemplar paths from a repository of exemplar paths each representative of a previously characterised craft movement path, and determine a path match likelihood for each candidate exemplar path.

The pattern matcher can select a first craft position from the craft position data for use to search the repository of exemplar paths based on the first craft position to identify any exemplar paths proximate the first craft position, and

for each identified proximate exemplar path:

determines relative heading and relative path distance between the exemplar path and craft movement path; and

assesses whether or not to select the exemplar path as a candidate path based on candidate criteria.

In an embodiment the relative heading between the exemplar path and craft movement path is determined by selecting at least one further craft position from the craft position data, determining the distance between each further craft position and a position of the exemplar path which corresponds to the further craft position sample, and determining the relative heading based on difference in distance between exemplar path and craft movement path at the first vessel position and further vessel position.

In an alternative embodiment the relative heading between the exemplar path and craft movement path is determined by comparing an absolute heading for the craft movement path obtained from the craft movement path data at the first position and an absolute heading for the exemplar path at a corresponding position.

An embodiment if the system further comprises a pattern miner adapted to characterise exemplar paths from historical craft movement data and populate the exemplar path repository with exemplar path data.

In an embodiment the pattern miner analyses historical data representing a plurality of tacked craft paths, identifies common sequences of movement between regions, identifies clusters of movement paths for each sequence, and for each identified cluster analyses the cluster to determine a characterising exemplar path for the cluster comprising a characterising craft movement path and craft distribution about the characterising craft movement path for the cluster.

I n an embodiment the pattern miner analyses a cluster to characterise an exemplar path by:

dividing each movement path of the cluster into a sequence of equidistant segments positions along the movement path; and

I n an embodiment the characterising craft position and distribution are determined based on root mean square analysis of the distance between each movement path.

An embodiment of the system further comprises an exemplar path repository in the form of a database storing exemplar path data searchable using geospatial database queries.

An embodiment of the system comprises a pattern matcher controller and two or more pattern matchers in data communication with the pattern matcher controller and each pattern matcher executable using separate processing hardware resources and wherein the pattern matcher controller is adapted to allocate one or more sequences of craft position data samples to each pattern matcher for identification of candidate exemplar paths.

According to another aspect of the system there is provided a pattern matcher adapted to identify, based on one or more craft position data samples, any one or more candidate exemplar paths from a repository of exemplar paths each representative of a previously characterised craft movement path by:

searching the repository of exemplar paths based on a selected first craft position to identify any exemplar paths proximate the first craft position;

determining relative heading and relative path distance between each exemplar path and craft movement path;

assessing whether or not to select each exemplar path as a candidate path based on candidate criteria; and

determining a path match likelihood for each candidate exemplar path.

An embodiment of the pattern matcher can determine the relative heading between an exemplar path and craft movement path by selecting at least one further craft position from the craft position data, determining the distance between each further craft position and a position of the exemplar path which corresponds to the further craft position sample, and determining the relative heading based on difference in distance between exemplar path and craft movement path at the first vessel position and further vessel position.

An alternative embodiment of the pattern matcher determines the relative heading between the exemplar path and craft movement path by comparing an absolute heading for the craft movement path obtained from the craft movement path data at the first position and an absolute heading for the exemplar path at a

corresponding position.

A pattern matcher as described above can be implemented using a plurality of distributed data processing resources.

According to another aspect of the present invention there is provided a pattern miner adapted to characterise exemplar paths from historical craft movement data by: analysing historical data representing a plurality of tacked craft paths;

identifying common sequences of movement between regions;

identifying clusters of movement paths for each sequence;

analysing each cluster to determine a characterising exemplar path for the cluster comprising a characterising craft movement path and craft distribution about the characterising craft movement path for the cluster; and

populating an exemplar path repository with exemplar path data.

In an embodiment the pattern miner can analyse a cluster to characterise an exemplar path by:

An embodiment of the pattern miner as described above can be implemented using a plurality of distributed data processing resources.

According to another aspect of the present invention there is provided computer program code which when executed causes a computer to implement a craft motion path prediction method as described above. Brief description of the drawings

An embodiment, incorporating all aspects of the invention, will now be described by way of example only with reference to the accompanying drawings in which

Figure 1 is a block diagram of an embodiment of a system to aid craft movement prediction;

Figure 2 is a flowchart illustrating an example of a method predicting a movement path for a tracked craft; Figures 3a-c illustrate an example of a comparing exemplar paths to a movement path of a tracked craft;

Figure 4 is a flowchart illustrating an example of a process for characterising exemplar paths;

Figures 5a-d illustrate examples of charactering an exemplar path;

Figure 6 is a flowchart illustrating an example of a process for characterising an exemplar path for a cluster;

Figure 7 is a block diagram illustrating an alternative embodiment of the system to aid craft movement prediction;

Figure 8 is a block diagram of a further alternative embodiment of a system to aid craft movement prediction.

Detailed description

Embodiments of the present invention provide a method and system for aiding the prediction of craft movement paths by obtaining one or more craft position data samples for a craft and identifying, based on the craft position data samples, any one or more candidate exemplar representative of a previously characterised craft movement path, and determining a path match likelihood for each candidate exemplar path.

In the context of this specification craft is used to refer to any type of manned or unmanned land, airborne or waterborne craft. Airborne craft include any aircraft, for example airplanes, helicopters, airborne weapons, drones, missiles, balloons etc. Waterborne craft include any type of watercraft for example ships, boats, yachts, hovercraft and other vessels etc. Land craft include any land vehicles for example, trucks, cars, tanks, motorcycles etc.

An exemplar path characterises a craft movement path through a sequence of regions and distribution for craft following the movement path. In simple terms, an exemplar path represents a line through space representative of a craft movement path and an envelope for craft distribution about this line.

Exemplar paths are characterised by analysing historical craft path data to determine patterns of movement paths. Individual patterns or clusters of movement paths are further analysed to determine a characterising craft movement path and craft distribution about the characterising craft movement path for each cluster.

Exemplar paths can be defined using a series of spatial data elements representative of a characterised craft movement path. Each spatial data element can comprise data indicating a position in space and a distribution associated with the position for craft following the path. The spatial data elements include two dimensional Cartesian coordinates. These coordinates can be latitude and longitude coordinates. However, various embodiments position coordinates may be defined differently, for example using distance with reference to a specified position. In some embodiments craft positions may be defined in three dimensions, for example using latitude, longitude and altitude. Three dimensional exemplar path data may be desirable where the craft of interest are aircraft or other airborne craft such as weapons. However, airborne craft can equally be tracked using only two dimensional, latitude and longitude, position data. Some current radar technology is inaccurate when measuring altitude. Whether or not an embodiment includes altitude information in exemplar track spatial data may be based on the accuracy of the altitude data available for exemplar track characterisation. For example, exemplar paths may be characterised in two dimensions, latitude and longitude, only until technology capable of providing accurate altitude data is used widely enough for reliable altitude data to be included in the data used for characterisation of the exemplar paths. However, even where accurate altitude data is available some embodiments may characterise exemplar paths in only two dimensions, latitude and longitude. An advantage of using only two dimensions is reduced processing capacity is required for path matching

In some embodiments planned movement paths such as defined flight paths or shipping lanes may also be defined in terms of exemplar paths.

Figure 1 is a simplified block diagram of an embodiment of a system 100 for aiding craft movement path prediction. The system 100 comprises a path monitor 1 10 and a pattern matcher 120. The path monitor 1 10 is adapted to obtain craft movement path data for at least one craft. For example, in an embodiment where the system is used for aiding craft path prediction in an air space, the path monitor 1 10 can be in data communication with one or more real time aircraft tracking systems such as civilian and military radar, satellites, air traffic controllers, identification friend or foe (IFF) systems etc. The path monitor 1 10 can obtain real time aircraft path data 130 via these systems. The real time aircraft path data 130 may also include data indicating identified aircraft. For example, civilian and military air traffic controllers may provide flight numbers, flight plans, aircraft identification, aircraft type, IFF data etc. for identified aircraft. For example, all data typically used for building the recognised air picture (RAP) for the airspace may be provided to the path monitor. The path monitor 1 10 can be adapted to identify and select, out of the available real time data 130, data samples that relate selected craft. For example, selected craft may be unidentified craft, aircraft not following an anticipated flight plan, aircraft nominated for monitoring and path prediction by an official body etc. Alternatively, RAP data may be filtered such that only data samples relating to selected craft are obtained by the path monitor 1 10. For example, the path monitor may only be provided with sample data sets for unidentified craft or other selected craft.

Alternative embodiments may be applied in surface craft path monitoring and in such embodiments the path monitor may be in data communication with civilian and military surface level radar, satellites, coast guard, traffic control, shipping controllers etc. Embodiments may also monitor both surface and airborne craft. For example, an area of interest may be an area of ocean where both shipping and airborne traffic are of interest. In an alternative example, all land, sea and air traffic may be of interest around a strategic naval base.

The path monitor 1 10 obtains craft position data samples for at least one craft, for example, a series of craft positions detected by periodic radar scanning.

The path monitor 1 10 can be adapted to obtain craft path data including real time craft path data directly or indirectly from a capture source. For example, capture sources can include primary or secondary radar or air traffic control etc. The path monitor can be adapted to transform the obtained data into a series of craft position data samples for use by the pattern matcher 120. For example, such a transformation may include data format conversion, data sampling, interpolation or extrapolation of captured data, sampling rate conversion etc. The transformation may also include combining craft position data obtained from two or more sources into a single series of craft position samples. For example, craft position data obtained from two radar sources or radar and satellite sources may be combined to manipulate the obtained data into a single set of craft position samples suitable for use by the pattern matcher 120.

The path monitor 1 10 may also be adapted to perform faded track correlation where some data samples may be lost for a tracked craft. For example, in an area where radar coverage is scant or affected due to interference, craft position data may be received irregularly resulting in apparent "breaks" in the tracked movement path. The path monitor 1 10 can be adapted to extrapolate two or more sections of movement path to determine whether or not the sections appear to originate from the same craft. For example, two sets of only a few position data samples can be compared to determine whether or not the headings for the two sets of samples are similar. Extrapolation between the end of one set of data samples and the beginning of the next can indicate whether or not the two sets follow substantially the same path and are likely to relate to the same tracked craft. The path may be a straight line but other patterns typical of movement paths may also be extrapolated. For example, an arc or turn may be extrapolated where the two sets of position data samples have different headings. For example, two sets of position data samples each ascribing an arc of a similar radius may enable extrapolation of a section of arc between the two sets of data samples.

Information additional to the heading of the two sets of samples can be used to determine whether or not the two set of tracked data relate to the same craft. For example, the aircraft's average speed may be compared for the two sections of tracked data. Further the time and distance between two sections of tracked data can be used to calculate the approximate speed the craft would have had to be travelling for the two sections to be part of the one movement path of the craft. This calculated approximate speed may be compared with the actual speed of the craft determined for each of the sets of position data samples. It may be assumed that the craft's speed should not chance dramatically between the two actual and one calculated value. Further, the nature of any variation may also be indicative of whether or not the two sections relate to the same craft. For example, where the craft is moving at a constant speed for both sections of the actual tracked path, it could be assumed that the craft will travel at a similar speed during the break in the tracked path. Where the difference between the calculated and actual speed is outside a threshold tolerance, it may be assumed that the two sections of tracked path relate to two different craft. Where the speed is different between actual and calculated path sections, the difference in speed can be analysed to determine whether the speed changes could be indicative of a sign craft's regular movement patters. For example, do the differences indicate the craft is smoothly accelerating?

Where two sections of path are identified as relating to the same tracked vessel, the path monitor can derive approximate position data samples between the sections of actual tracked data samples for use in exemplar path matching.

Faded path correlation has an advantage of improving input data quality for exemplar path pattern matching. In an example two sets of vessel positions obtained through the real time tracking systems, each only having two or three data samples. In this example, if faded path correlation indicates the two sets of data samples appear to be from the one craft, this means more actual position data samples are available for pattern matching to exemplar paths. Further, if data samples can be interpolated between the two sections of actual data samples, the set of craft position data samples for use in pattern matching can be further increased. In turn, having more data samples for matching to an exemplar path can improve the likelihood of identifying the correct exemplar path.

Typically faded track correlation is performed by the path monitor prior to pattern matching of position data samples for a tracked craft with exemplar paths. However, in some embodiments faded track correlation may be performed after pattern matching with an exemplar path in order to determine whether a newly identified set of position data samples relates to a previously identified craft or a different craft potentially following the same exemplar path. For example, a live track may be correlated to a path of a previously identified craft and full exemplar path pattern matching avoided.

The path monitor 1 10 may be adapted to provide a series of a given number of craft position data samples, representative of the craft's path over a given distance. For example, ten samples over a travel distance of five kilometres may be specified. Each position may be equidistant along the craft's movement path. However, this is not essential and the distance between craft positions can be varied. An advantage of using equidistant samples is that samples from the craft path and corresponding positions of an exemplar path can be accurately correlated and this correlation is independent of time. This can simplify the comparison and reduce the amount of processing required.

Using equidistant craft positions can also have an advantage of providing more consistent pattern matching results between different areas where the data capture performance varies. For example, where radar performance varies between different areas, using equidistant craft position samples can have an effect of compensating for the inaccuracies and improving the consistency of the pattern matching outcomes. Data capture performance variation has many causes for example, variations in radar sampling rates and resolution, atmospheric or electromagnetic interference, capability variation between different technologies used for capture and processing of real time data, etc.

The series of data samples obtained by the path monitor 1 10 is used by the pattern matcher 120 for identifying candidate exemplar paths which may be the path being followed by the tracked craft. The pattern matcher 120 is in data communication with an exemplar path repository 140. Based on one or more craft position data samples the pattern matcher 120 identifies any one or more candidate exemplar paths from a repository of exemplar paths. In some instances no candidate exemplar paths will be identified. For example, the tracked craft may not be following a route characterised by and exemplar path. Alternatively an aircraft may have had to deviate from a planned flight path, for example to avoid poor weather conditions, and this deviation taken the path outside a tolerance distance for what would otherwise be an exemplar path for the planned flight path.

Where data samples for the tracked craft indicate the craft's heading an exemplar path match may be determined based on one position data sample alone. For example, some radar and satellite tracking systems may provide position data samples defining an absolute position and heading for the craft. For example, longitude and latitude coordinates and also a heading may be provided in terms of degrees, minutes and seconds, depending on the sophistication and accuracy of the equipment. Alternatively, heading can be determined by comparing the difference in position between two data samples. For example, where only latitude and longitude coordinates are provided heading can be calculated based on position difference between sequential data samples.

The pattern matcher 130 is also adapted to determine path match likelihood for each candidate exemplar path. Each exemplar path is defined using a sequence of spatial data elements representative of a previously characterised craft movement path. The characterisation of craft movements and definition of exemplar paths may be performed prior using historical data of craft movements in the airspace. Each exemplar path may be characterised using a sequence of waypoints and headings or alternatively a series of position coordinates.

The pattern matcher 130 is in data communication with the exemplar path repository 140 to obtain exemplar path data. For example the exemplar path repository 140 may be a database accessible to the pattern matcher 130 for example via a local area network connection or via a communication network and server.

Although a database is used in this example for the exemplar path repository, any feasible computer readable memory and data structure may be used to implement the repository. For example, index, library, array or other data structures may be used for storing exemplar path data. The repository may be stored on a server or hard drive, alternatively the repository may be stored on transportable computer readable medium such as a optical disc, solid state or flash memory etc. It should be appreciated that any available technology either currently available or available in the future could be used.

In some embodiments the exemplar path repository 140 may be part of the system 100 and may be integrated with the pattern matcher 130. For example, the exemplar path repository 140 may be a database of exemplar paths stored in memory of a data processing system in which the pattern matcher 120 and path monitor 1 10 are also implemented. For example, the path monitor 1 10 and pattern matcher 120 may be implemented as software modules executable using a processor which accesses memory storing the exemplar path repository 140 in a database structure. Herein the term "processor" is used generically to refer to any device that can process instructions and may include: a microprocessor, microcontroller, programmable logic device or other computational device, a general purpose computer (e.g. a PC) or a server. Alternatively, the path monitor 1 10 and pattern matcher 120 may be

implemented using a combination of hardware, software and firmware and may utilise a combination of shared and dedicated data processing hardware and memory resources. For example, a dedicated hardware circuit, such as an application specific integrated circuit (ASIC), may be used to implement some or all of the pattern matcher and/or path monitor functionality. This hardware circuit may be used in a data processing system having at least one processor, memory and other resources for executing cooperating firmware and software to support the full functionality of the pattern matcher and the path monitor and integrate with external systems such as an exemplar path database and craft path data capture systems. It should be appreciated that many alternative system architectures could be used to implement the system and all such alternatives are envisaged within the scope of the present application.

To identify candidate exemplar paths, the pattern matcher 120 selects a first craft position from the craft position data. The repository 140 of exemplar paths is searched to identify any exemplar paths proximate the first craft position. For each identified proximate exemplar path the relative distance and heading between the craft path and exemplar path is compared. Where heading data is not included in the sample data of each vessel position, the pattern matcher 120 uses two or more positions to determine the relative heading. The pattern matcher 120 selects at least one further craft position from the craft position data. The pattern matcher 120 determines the distance between each further craft position and a corresponding position in the exemplar path. The corresponding position in the exemplar path is a position in the exemplar path the same distance along the exemplar path as the distance between the first and second positions of the craft path. Candidate criteria are then applied to asses whether or not to select the exemplar path as a candidate path. Comparison of vessel and corresponding exemplar path positions can be repeated for as many craft position data samples as is required to determine whether or not to exclude the exemplar path as a candidate path. For example, an exemplar path may be excluded based on comparison of one or two positions where the exemplar and craft paths are heading in opposite directions. Whereas where the craft path has a similar heading to the exemplar path all available position data samples may be compared.

An example of a pattern matching process to identify candidate exemplar paths is illustrated in Figure 2 with reference to Figures 3a-c. The process 200 starts with obtaining a series of craft movement positions 210. A first craft position is selected from the craft position data samples for searching the exemplar path repository for proximate paths 220. The search of the exemplar path repository will return all exemplar paths which pass proximate the first craft position. For example, exemplar paths which pass within a given distance of the first craft position are returned from the search.

Any position in the craft position sequence may be selected. However, using the most recent craft position can have an advantage of minimising the number of incorrect exemplar paths returned from the search. For example, where an earlier craft position is selected as the first position, an exemplar path may be returned from the search which the craft has since moved away from and is therefore obsolete as a candidate exemplar path. For example, a path that was intersected rather than followed. Further, an exemplar path that the craft is approaching and potentially following may miss being included in the search result, for example if the craft has recently changed direction.

In the illustrated process the most recent craft position is selected as the first craft position. The search 220 of the exemplar path repository returns one or more paths proximate the first craft position.

In an embodiment the exemplar path repository is a data base of exemplar path data searchable using geospatial database queries. For example the search can use geospatial database queries to search for exemplar paths that pass within a given distance of the first craft position. For example, exemplar paths having a

characterising craft movement path passing within 10 km of the first craft position are returned from the geospatial database search. These exemplar tracks may be returned ordered based on distance form the first craft position, for example closest first.

An optional second stage can be applied to the search where the distance between the characterising craft movement path and first craft position is compared with the characterised craft distribution, for example the standard deviation for craft distribution, about the characterising craft movement path proximate the first craft position. Exemplar path candidate criteria may define a distance threshold in terms of standard deviation, for example three standard deviations. A distance based on standard deviation is used based on an assumption of a normal distribution of movement paths around a characterising exemplar path and therefore any craft following the exemplar path would typically be found within three standard deviations of the characterising path. Exemplar paths may be excluded where the distance between the characterising craft movement path and first craft position is greater than three standard deviations of the characterising craft movement path. It should be

appreciated that the standard deviation for each exemplar path may be different and the standard deviation may also vary along the path. Where a standard deviation envelope varies along an exemplar path, and average of the standard deviation distance may be used for determining a threshold distance to use for the initial search. Alternatively, a standard deviation value for each position along the exemplar path may be used. A measure other than standard deviation may be used to indicate the distribution of craft movements around a characterising path, for example a maximum outlier distance may be defined for each exemplar path. The first craft position may need to be within a multiple of this distance to avoid the path being excluded, for example one and a half times the maximum outlier distance. Any such measures and values or functions thereof may be used for searching exemplar paths.

The accuracy of the search may also be altered, for example by reducing the distance or multipliers for standard deviation or outlier distance, depending on the number of exemplar paths identified in the first stage of the search. For example, where one hundred exemplar paths are returned for the initial search, this may be reduced to the fifty closest paths based on proximity in the first stage. The second stage may compare these closest fifty exemplar paths to exclude those where the first craft position is outside three standard deviations. If all fifty paths remain, a further comparison using a multiplier of two and a half standard deviations may then be used to reduce the number of exemplar paths to around twenty to thirty paths for further processing. Alternatively, the closest fifty may be ranked again based on multiplier of standard deviation for the distance between the first craft position and the

characterising path. This may alter the ranking order from an order simply based on distance. The top thirty exemplar paths may then be selected for further processing. This may result in paths where the first craft position is within two standard deviations being selected. In this case the accuracy of the search is limited by the number threshold rather than any defined distance or standard deviation threshold.

For example, with reference to Figure 3a the first craft position 310 lies near two separate intersecting characterising craft movement paths 320 and 340. Each characterising craft movement path 320 and 340 has an associated distribution envelope, indicated in Figure 3a by dotted lines 322 and 325 around path 320 and dotted lines 342 and 345 around path 340. This envelope may be indicative a multiplier of the standard deviation for the path 320, 340 or a given distance. The distance di 330 between the craft position 310 and path 340 is further than the distance d₂ 370 between the craft position 310 and the path 320.

Once the search has identified proximate exemplar paths 220, an exemplar path is chosen for further comparison 230. In the embodiment illustrated the first exemplar path chosen 320 is the path closest to the craft. However, any path may be chosen. An advantage of starting from the closest path is that it is anticipated that typically a path match is more likely to be located with in the closest proximity paths.

One or more further craft positions can be compared to corresponding positions on the characterising path of the chosen exemplar path 240. For example, as illustrated with reference to Figure 3b a second craft position, in this case the next previous position 350, is compared with a corresponding position on the characterising movement path 320. The position on the characterising path of the exemplar path closest to the first craft position is the first corresponding position for the exemplar path. The distance between the first and further craft positions on the tracked movement path is used to determine the corresponding position on the exemplar path. Each further corresponding position for the exemplar path is a position the same distance along the characterising path from the first corresponding position as the distance between the first and further positions on the tracked movement path. For example, where the first and second tracked positions are one kilometre apart, the corresponding positions along the characterising path of the exemplar path will be one kilometre apart.

The distance d₃ 370 between the further craft position 350 and path is measured 245 and a heading difference determined. The two craft positions 310 and 350 can be extrapolated to give a craft heading 365 for the craft and a difference angle Θ between the craft heading 365 and the exemplar path heading 360 can be calculated. The heading difference can also be used to automatically exclude 250 exemplar paths which travel in the opposite direction to the direction of travel for the craft.

Where heading data is not provided with samples position data, a minimum of two craft position samples are required in order to determine the craft heading.

However, a number N craft position samples may be compared with the exemplar path. For example, N can be chosen to include only a recent portion of the craft path, for example five samples. In an embodiment the number of samples is chosen based on a distance of travel for path matching. The distance can be based on the sample rate or independent of the sample rate for the craft position.

For example, where a craft is travelling at a constant velocity and craft position samples are taken at a constant sample rate the distance D travelled by the craft between samples will be constant, for example 1 km. In this case a set of craft positions may be provided directly from the real time sampled data, for example from a radar output. However, in reality distance travelled will typically not be constant between real time data samples. Processing of the real time sampled data can be performed to interpolate the craft path and derive a set of craft positions which are equidistant along the path travelled. For example, the path monitor 1 10 can perform this interpolation and provide the required number of craft position samples, say five samples each one kilometre apart giving the most recent five kilometre section of the craft's path for pattern matching.

The number N of craft position data samples chosen for comparison may be based on a given distance D, for example five samples each 1 km apart for a total distance D of 5km. The distance, sample rate, and number of samples can vary between tracked craft and systems. Where the tracked craft data is available, a number of recent samples N, where N is greater than one, can be chosen to optimise accuracy of path matching and processing time. Alternatively, where data representing only a very small section of a craft's path is available, say one or two kilometres, then the distance D may be chosen based on the available data.

Each of the craft positions can be compared to a corresponding position chosen on the characterising craft path of the chosen exemplar path. The corresponding position on the exemplar path can be determined based on the distance between craft position samples and the direction of travel of the exemplar path. It should be appreciated that it is not essential to use equidistant vessel positions. However, using equidistant vessel positions can have an advantage of reducing processing required for pattern matching.

Where more than two craft positions are used the average heading difference and distance between the tracked craft path and exemplar path can be calculated. Alternatively the distance and heading difference can be compared for each craft position and these differences compared to identify trends for the sequence. For example, where the distance between the tracked craft and the exemplar path is changing greatly at a relatively constant rate between craft positions this can indicate that the tracked craft path intersects, rather than follows the exemplar path, and the exemplar path may be excluded from consideration based on the rate of change or angle of intersection between the tracked craft path and exemplar path. For example, as illustrated with reference to Figure 3c the second craft position 350 is compared with a corresponding position 341 on the characterising craft movement path 340. The distance d₄ 335 between the further craft position 350 and path is measured 245. The distance between the craft path and the exemplar path varies significantly between position 310 and 350, it can also be seen in Figure 3c that the craft heading 365 intersects the exemplar path 340. This difference in path direction may be outside a range specified in candidate criteria and the exemplar path 340 excluded as an exemplar path.

Where the relative distance between the tracked craft and exemplar path remains substantially constant the exemplar path may be selected as a candidate exemplar path potentially being followed by the tracked craft. For example, as illustrated in Figure 3c the path of the craft 360 and exemplar path 320 are similar, so exemplar path 320 may be selected as a candidate exemplar path, whereas exemplar path 340 is excluded.

Based on the distance and heading difference, a decision is made 250 whether or not to exclude an exemplar path from consideration as a candidate exemplar path. Where an exemplar path is excluded 250, the next closest exemplar path is chosen 230 for further comparison and the process above is repeated.

Where an exemplar path remains chosen as a candidate exemplar path 250 the probability of a match to the craft track path is calculated 260. This calculation can use a combination of the distance between the tracked craft path and characterising movement path of the exemplar path and heading difference. For example, a component of probability based on distance can be calculated based on the Cartesian distance between the tracked craft positions and corresponding positions on the characterising craft movement path of the exemplar path and the standard deviation for the exemplar path. This probability may be calculated from an average of the number N of craft position samples used.

For example for a given tracked craft position, Nt, the corresponding exemplar path position, Ne, and standard deviation, δ, can be used to calculate a match probability. The Cartesian distance, d, between Nt and Ne is determined. Distance d is converted to z assuming a standard normal distribution.

d

z =— Equation [1]

σ

Using the area under the standard normal distribution in the ranges -infinity to -z and +z to + infinity gives the properties:

d close to mean => P=1

d far from mean => P=0 Equation [2]

where P is the probability of matching.

The probability of matching, P, is given by:

p = l - e^ Equation [3]

The probability, P„ can be calculated using the position difference, of,-, for each position and averaged to give an overall probability based on distance.

i_Qd_i Equation [4]

A component of probability based on heading distance can be calculated from the difference between the heading of the tracked craft and exemplar path, again this can be calculated from an average for the N craft position samples.

For example, exemplar paths having a heading difference greater than a maximum candidate criteria threshold, for example 30 degrees, have been excluded prior to the probability calculation. Heading probability has the properties:

heading difference of 0 => P=1

heading difference > maximum difference => P=0 Equation [5]

The heading difference based probability can be given by:

pheadin_g = ¹ sheading Equation [6]

max dijjerence

The distance and heading probabilities can be combined to produce an overall probability.

^match ^dist ^heading * w heading Equation [7]

where w_dist and w_headin_g are weightings with

^dist ^heading ^~ ^

Any weighting may be applied depending on the embodiment. However, typically a higher weighting is given to the distance probability that the heading probability. For example:

^dis, = 0-75

Equation [8]

^heading = 0·²5

The candidate exemplar path is then ranked 265 for probability of match relative to other candidate exemplar paths. Once all exemplar paths have been compared and ranked in this manner 270, the outcome is presented to an operator 290. In embodiments the list of exemplar paths and their match likelihood are displayed to an operator. The operator may selectively view each candidate exemplar path to choose which appears to be the most likely match 290. In some embodiments the pattern matcher may be adapted to automatically select the highest ranking candidate path to display to the operator in the first instance. The operator may then selectively view other candidate exemplar paths. If more than one exemplar path has the same highest probability rating, all such paths may be displayed to the operator to enable the operator to choose a path match.

After a tracked movement path and exemplar path have been matched, the path monitor continues to obtain position data for the craft. This position data is periodically compared to the matched exemplar path. The comparison is performed similarly to the pattern matching described above, but for the matched exemplar path only. The periodic comparison is performed in order to identify any deviation from the exemplar path that may indicate either the exemplar path was incorrectly identified as a match or that the craft has changed course. Where a deviation from the matched exemplar path is identified the operator can be alerted. In response to the alert the operator may investigate the deviation. For example, this may include revisiting the previous candidate exemplar paths to identify an alternative match or triggering a new pattern matching procedure. Alternatively, other data for the tracked craft may be reviewed by the operator, for example where an aircraft matched to an exemplar path characterising a commercial passenger airline route deviates from the exemplar path and accelerates to mach 3 the operator may be instantly alerted that this aircraft appears to be military in nature.

Periodically re-checking correlation between tracked craft paths and exemplar paths to automatically identify deviations can reduce the risk of such deviations being overlooked. For example, RAP operators can be required to oversee and identify multiple aircraft every minute. Once an aircraft has been identified, and particularly identified as non-threatening, it is easy for an operator not to notice a deviation from the anticipated flight path, or only identify a deviation when the aircraft is a long way off course. This often happens quite simply because of the amount of data the operator has to take in and process manually at any instant. Automatic periodic re-checking of path matched can help reduce the load on operators and minimise the risk of a path deviation being overlooked.

Exemplar paths are characterised based on historic data. For example, historic data may be tracked fight path data of a region of airspace over a two month period. The amount of historic data used for characterising exemplar paths may be based on data availability and pattern miner processing capacity. For example, exemplar paths for a high traffic region of airspace may be able to be characterised based on one or two months tracked flight data. Whereas, tracked craft data for a period of several months or years may be used for characterising a low traffic region of airspace.

Exemplar paths may be updated periodically for regions of airspace. For example, periodically a region may be re-characterised so new exemplar paths can be added to a repository of exemplar paths for the region and obsolete exemplar paths removed or archived. Further, existing exemplar paths can be adjusted, for example where a typical flight path has changed a corresponding change may also be made in the exemplar path. This can be done by re-analysing the cluster of paths.

An example of a process of characterising exemplar paths is illustrated in

Figure 4. The process starts with analysing historical data representing a plurality of tacked craft paths 410 to identify common sequences of movement between regions. For example, the historical data of individual craft paths is analysed to identify regions passed through by more than one path and sequences of regions. For example, as illustrated in Figure 5a, paths 541-545 all originate at region A 510, and terminate at region C 530 and paths 541-544 pass through intermediate region B 520. Thus, a common sequence A-B-C for paths 541-544 may be identified from these tracked paths.

In an embodiment, historical tracked craft paths are subjected to path simplification using a version of a Douglas-Peucker algorithm for smoothing of polylines, as described in the article: David Douglas & Thomas Peucker, "Algorithms for the reduction of the number of points required to represent a digitized line or its caricature", The Canadian Cartographer 10(2), 1 12-122 (1973).

The Douglas-Peuker algorithm is a recursive algorithm which can be applied to simplify a curve made up of a plurality of line segments to provide a similar curve having fewer line segments, thus a simplified version of the curve. For example, path simplification using the Douglas-Peuker algorithm can result in a curved flight path being simplified into a sequence of way points where the aircraft has made a significant heading change and straight lines in between these way points. The reason for applying this algorithm is to reduce processing capacity required for characterising the exemplar paths. In some embodiments, where time and processing capacity is not a constraint, this path simplification step may be omitted.

The airspace is divided into regions. Common patterns of movement between the regions can then be identified. For example, closed sequential pattern mining (clospan) can be applied to identify common sequences of movement between regions. Clospan is an algorithm for mining closed sequential patterns, as described in the article: X. Yan, J. Han, and R. Afshar, "Clospan: Mining closed sequential patterns in large datasets," In SDM, 2003, pp. 166-177. The Clospan algorithm is an extension of a PrefixSpan algorithm that mines maximal patterns only, as described in J. Pei, J. Han, B. Mortazavi-Asl, H. Pinto, Q. Chen, U. Dayal, and M.-C. Hsu, "PrefixSpan:

Mining Sequential patterns efficiently by prefix-projected pattern growth," Proc. 2001 Int. Conf. Data Engineering (ICDE'01), pp 18-25, McLean, VA, Nov. 2002.

Mining maximal patterns is a method for removing redundant sequences. For example, if two common sequences are discovered A->B supported by four paths and A->B->C also supported by four paths, only the longer (maximal) sequence is reported by Clospan. Although Clospan is given as example of a pattern mining algorithm for indentifying common sequences any suitable algorithm may be used. It should be appreciated that identification of sequences based on movement through regions means order is inherent in the sequence. This avoids paths passing through the same regions but in opposite directions being identified as a common sequence.

Each sequence can then be analysed to identify clusters of movement paths for the sequence. An identified common sequence is selected 425 and analysed to identify clusters of movement paths for the sequence 430. There may be more than one cluster per sequence. For example, two or more different routes may pass through the same sequence of regions. In an embodiment, shared nearest neighbour (SNN) density based clustering is used to identify groups of similar paths following the sequence. Shared nearest neighbour clustering is described in the article: R.A. Jarvis and Edward A. Patrick, "Clustering Using a Similarity Measure Based on Shared Near Neighbours," IEEE Transactions on Computers, Vol. C-22, No. 1 1 , November 1973. However, other algorithms may be used to identify clusters.

For example, as illustrated in Figure 5a and b, SNN density based clustering may identify a cluster formed of paths 542-544. In this example, paths 541 and 545 have been removed through application of the SNN clustering algorithm. Paths, 541 and 545 are outliers which do not form part of the cluster 542-544.

A cluster is selected 435 and analysed to characterise the exemplar path for the cluster 440. Characterising the exemplar path comprises determining a characterising craft movement path and craft distribution about the characterising craft movement path for the cluster. The exemplar path data is then stored in an exemplar path repository 450. The process can them be repeated for the next cluster 460 or next sequence 465 until all sequences have been characterised.

An embodiment of a method for characterising an exemplar path for a cluster is illustrated in Figure 6. The process 600 starts 610 by dividing each movement path into a number X of equidistant segments. Each movement path is divided into the same number of segments X. This enables the segment length to vary between paths to compensate for variations in path length due to different routes. For example, as illustrated in equation 1 each path of the cluster of N paths can be represented by a sequence P of positions p.

= ^/½,^/½,^/¾ -^/^ Equation [9]

Further, using the same number of samples for each path of a cluster enables the samples to be time independent, which has an advantage of simplifying the analysis for the cluster.

The first, m=1 , segment, S_m, for each cluster is selected 625.

S_m = p_lm , p_2m , p_3m ...p_nm where \ < m < x Equation [10]

The distribution for each segment S_m can be analysed to determine a characterising craft position for the distribution 640. For example, the characterising craft position p_em for the segment S_m can be determined based on a root mean square analysis of the distribution of the positions p_1m to p_nm. The root mean square point between the positions p_im to p_nm can be used as the characterising position for the segment p_em for the segment S_m. The standard deviation 5_em for the distribution of the segment S_m can also be calculated 650. The standard deviation can be used to create a "window", say a three standard deviation window, around the characterising craft

5 path where craft following the route characterised by the exemplar path would be

expected to be found.

The characterising position p_em and standard deviation 5_em can be stored 660 as a spatial data element e_em for the exemplar path £.

E = e_l , e₂ ,e₃•••e_x Equation [1 1] i o where

1 < m < x

^E = {Pel , °el {Pel , ° el {P , ° ^' ei )" " " {P ex , ° ex ) Equate [12]

15 Exemplar path data can be two dimensional, for example Cartesian coordinates or latitude and longitude coordinates. Performing exemplar path characterisation in only two dimensions can have the advantage of simplifying processing, thus, reducing required processing capacity. Further, current tracking technologies commonly in use either do not provide altitude data or do not provided altitude data with sufficient

20 accuracy for reliable pattern matching. However, performing exemplar path

characterisation having three dimensional position data is also envisaged within the scope of the present application.

The analysis repeated for the next segment 675, m=m+1 , until the last segment m=x has been analysed 670. The spatial data for each segment of the exemplar path

25 comprises a characterising craft position and standard deviation. Interpolation

between segments can be used to derive a characterising craft position anywhere along the exemplar path. Further, the standard deviation can also be interpolated between samples to determine the standard deviation at any position along the exemplar path. This can be used to create a window or envelope along the exemplar

30 path, for example a three standard deviation window. For example, a characterising craft path 550, 560 is illustrated in Figure 5c showing the exemplar path for route A-B- C 510-520-530. The envelope boundary of a three standard deviation 570 window about the characterising path is illustrated using dotted lines 572, 575. It should be appreciated that the boundary of this envelope may vary along the exemplar path, in

35 accordance with the distribution of the characterised cluster.

Once each segment of the exemplar path has been characterised 670 any required post processing of the exemplar path data can be performed. For example, in some embodiments each element of the exemplar path may include heading data indicating a direction or vector between characterising positions. The heading data may be calculated by determining the direction between adjacent characterising craft positions of the exemplar path. Another example of post processing may include smoothing of the distribution envelope about the exemplar path. Further

characterisation of the exemplar path may also be performed during post processing. For example, where altitude data is available characteristics of high, medium and low altitude aircraft for an exemplar path may be identified from the original path data and stored associated with the exemplar path data. For example, typical altitude, air speed or aircraft type may be characterised. Such data may be searched if desired by the operator. Such data may also be cross checked against a tracked craft. For example, an operator may be alerted to an aircraft flying above a maximum altitude possible for a passenger aircraft but following an exemplar path characteristic of a passenger flight.

Post processing can also include storing data for all craft paths used to characterise an exemplar path. This craft path data can be associated with the exemplar path to enable the original craft path cluster data to be retrieved. For example, if it is desired to re-characterise the exemplar path with greater accuracy. Alternatively an exemplar path may be re-characterised if further craft path data becomes available, including circumstances where a new tracked craft path is matched to the exemplar path. For example, an embodiment may be adapted to dynamically update the exemplar path data based on confirmed matches of the exemplar path to live craft path data. However, typically exemplar paths will only be characterised periodically, say monthly, due to the high processing capacity required.

Original craft path data of a cluster for an exemplar path can also be used to recover time based information. For example, a cluster used to characterise an exemplar path for a domestic aircraft route in Australian airspace between Adelaide and Melbourne can be stored with time data as well as position data for each path. For example departure time for each flight may be used as a metric for searching a subset of the cluster. Alternatively, a tracked craft may be correlated with a particular flight number or departure time, for example a flight from Melbourne to Adelaide departing Melbourne at 9am, by searching the cluster data for the exemplar path based on time. Other information such as craft speed and altitude may also be retried.

In an alternative embodiment, exemplar path characterisation can be performed based on identified common sequences alone, rather than analysing clusters of paths. In this embodiment the characterising movement path for the exemplar path is determined using a line between the centroid of each region of the sequence. For example as illustrated in Figure 5d, the sequence of regions A-B-C 510-520-530 is identified. The first leg 580 of the characterising movement path is determined by joining the centroid of region A 510 with the centroid of region B 520. The second leg 585 of the characterising movement path is determined by joining the centroid of region B 520 with the centroid of region C 530. The distribution 592, 595 about the characterising path 580,585 can be based on a simple distance 590 based window rather than a standard deviation based window.

This embodiment provides a less accurate characterisation of the exemplar path than a cluster distribution based characterisation. However, this simple characterisation embodiment requires less processing than cluster based distribution characterisation. Further, this simple region based exemplar path characterisation may be applied if, for an application, an exemplar path characterisation is desired for a single path or outlier to another exemplar path cluster. For example, an outlier path will not form a cluster and hence be excluded from cluster based characterisation. In some embodiments, excluded outlier paths may be subject to region based

characterisation. Full cluster path data can be stored associated with the exemplar paths as described above. Being able to access the full original path data used to characterise the exemplar path may enable compensation for inaccuracies introduced through the simplification for characterisation.

In some embodiments specified minimum threshold number of paths must be present for a cluster to be recognised. For example, an embodiment may specify that support of a minimum of ten paths through a sequence of regions is required to identify a pattern. The number of paths required to support a pattern may vary depending on the region. For example, in a remote region only four paths may be required to support a pattern, whereas near a major city twenty or fifty paths may be required to support a pattern.

It should be appreciated that characterisation of exemplar paths is not time critical to the operation of the system. The exemplar path characterisation is able to be performed as pre-processing to the pattern matching. The system requires exemplar path data to be accessible for performing pattern matching. However, the

characterisation of exemplar paths can be performed prior to pattern matching. For example exemplar path characterisation can be performed off-line using batch processing. Exemplar path characterisation may also be performed by a different system to the system performing pattern matching, provided the resulting exemplar path data is stored in a repository accessible to the pattern matcher. This has an advantage of reducing the processing capacity of the system required during pattern matching.

Further aspects of the method will be apparent from the above description of the system. It will be appreciated that at least part of the method will be implemented digitally by a processor. Persons skilled in the art will also appreciate that the method could be embodied in program code. The program code could be supplied in a number of ways, for example on a tangible computer readable storage medium, such as a disc or a memory (for example, that could replace part of or be added to memory of a data processor) or as a data signal (for example, by transmitting it from a server). Persons skilled in the art, will appreciate that program code provides a series of instructions executable by the processor.

Figure 7 illustrates an embodiment of a system 700 for implementing pattern mining and pattern matching as described above. The system comprises a path monitor 710, pattern matcher 720 and pattern miner 750. The system is in data communication with real time craft data capture systems 730 and one or more operator interfaces 770.

An exemplar path database 740 and historical craft path database 760 are provided which are accessible to the system 700 or may be included as part of the system 700. For example, one or more of the historical path database 760 and exemplar path database 740 may be external to the system and accessible via a communication network. In some embodiments more than one historical path database 740 may also be used. In some embodiments more than one exemplar path database 740 may also be used. For example, separate historical path databases and exemplar path databases may be provided, each relating to a different region of airspace. For example, separate databases may be maintained for an Australian region, European region, pacific region, Antarctic region, USA region etc. There may be some overlap between regions and corresponding overlap in the stored exemplar and craft path data in the different databases.

For example, a central government air traffic control authority may maintain a database of all tracked flight data for its airspace, this may be made accessible via a private data communication network, such as a secure wide area network, or public data communication network such as the Internet. The system 700 can retrieve data from the historical path database 760 via the network. For example, where the system is administered by a defence organisation the system can access the historical path database 760 via the secure wide area network. Alternatively, the historical path database 760 may be implemented as part of the system 700. For example, the historical path database 760 may be maintained by a defence organisation as part of the system 700. This may be desirable where some of the historical data is recorded by secret military radar or satellites. In such an embodiment the historical path database 760 may be populated using data acquired from a combination of private and publicly accessible civilian craft tracking data sources and historic databases, and secret military craft tracking data sources and historic databases.

In an embodiment the exemplar path database 740 is implemented as part of the system. Alternatively the exemplar path database 740 can be accessible to the system via a data communication network. In an embodiment where the exemplar path database is accessed via a data communication network, low network latency for communication between the system 700 and the exemplar path database 740 is desirable. Any delay caused by communication via the data network for retrieval of searched data from the exemplar path database, causes a corresponding delay in providing the candidate exemplar path data to the operator.

For example, when the system is performing pattern matching of live craft movement paths in real time for use in a battle management system, a delay of a few milliseconds in receiving candidate exemplar path data may be critical. For an application where time taken to return candidate exemplar paths is critical, such as where the system 700 is linked to a battle management system, the exemplar path database 740 may be incorporated in the system. For example, the exemplar path database may be stored in memory of a data processor system used to implement the pattern matcher. Alternatively, the exemplar path database may be connected to the system via a high speed local area network. In such an embodiment, access to the exemplar database via the internet may be unacceptable as typical network latency may be too high. Further, causes of delays may be outside the control of the system operator and this may also be unacceptable. However, for alternative applications where time is not as critical, for example for commercial air space monitoring or research purposes, an internet connection to an exemplar path database may be acceptable. It should be appreciated that the architecture may vary between embodiments of the system.

The system 700 received tracked craft data 730 via a real time craft tracking sources, such as primary and secondary radar 780, air traffic controllers 782, IFF systems 784, satellites, over the horizon radar, etc. Any craft tracking source can be used to obtain live craft tracking data. This data may be received directly from the source, i.e. a radar source, or indirectly via an intermediate system, such as a recognised air picture (RAP) production system 786. Data such as flight plans and craft identification data may also be obtained.

The system can be connected to one or more operator interfaces 770. The connection to operator interfaces may be direct or indirect. For example, operator interfaces 770 may be connected to the system 700 via a LAN. Using the operator interface the operator may select a desired region of interest for matching of live and exemplar craft paths. Alternatively, the system 700 may be connected to a battle management system (BMS) or recognised air picture production (RAP) system which, in turn, is connected to the operator interfaces. In this embodiment, an interface between the BMS or RAP system and the pattern matching system 770 can be provided to integrate the two systems. For example, this interface may enable the pattern matcher 720 to present exemplar path data on the display 775 of the operator interface 770. The interface may also enable the BMS or RAP system to interpret operator input in respect of candidate exemplar paths. For example, selection of a candidate exemplar path can be interpreted by the BMS or RAP system and in response update the RAP with the identified path. Alternatively the pattern matching system may be implemented integral to the BMS or RAP system.

An embodiment utilising distributed processing system architecture is illustrated in Figure 8. As illustrated the system 800 can be connected to a plurality of operator interfaces 872, 874, 876. The system 800 is in data communication with real time data capture systems 830 and a historical path database 860. Similar to the systems discussed above the system 800 has a position monitor 810, exemplar path repository 840 and pattern miner 850. However, the system 800 has a pattern matcher controller 820 and a plurality of pattern matchers 825a-n that can operate in parallel. Each of the pattern matchers 825a-n can be implemented using independent data processing resources for performing pattern matching of tracked craft as described above with reference to Figure 2.

For example, the pattern matcher controller 820 may be implemented in a server connected, via an internal network or LAN, to a bank of processors each programmed as a pattern matcher 852a-n to perform pattern matching as described above. The pattern matcher controller 820 can be adapted to distribute pattern match requests selectively to each of the pattern matchers 852a-n. Each of the pattern matchers may be adapted to access the exemplar path database for searching of exemplar paths and retrieval of exemplar path data. Alternatively, access to the exemplar path database 840 may be scheduled through the pattern matcher controller 820. For example, in an embodiment where the exemplar path repository is not adapted to handle simultaneous search requests the pattern matcher controller may perform search scheduling. For example, each pattern matcher 852a-n may send an exemplar path search query to the pattern matcher controller 820 which, in turn, sends the request to the exemplar path repository 840.

Alternatively, the pattern matcher controller 820 may be adapted to request an initial exemplar path search to identify exemplar paths proximate a tracked craft path. In this embodiment each of the individual pattern matchers 852a-n are provided with proximate exemplar path data and tracked craft position data. The pattern matchers 852a-n are therefore simply adapted to compare each proximate exemplar path with the tracked craft data to identify candidate exemplar paths and calculate the match likelihood for each candidate exemplar path, for example steps 230 to 265 as described above with reference to Figure 2.

A different operator may be using each of the operator interfaces 872, 874, 876. Each operator can select a region of interest for the airspace, for example by drawing a polygon around a region of a map to define the area of interest. The operator's interface displays a map of the area of interest. The map can show the recognised air picture and also highlight any unidentified craft in the air space. Pattern matching for unidentified craft may be automatically triggered by the system or triggered in response to an operator request. For example, an operator may know that a number of craft, as yet unidentified in the RAP displayed, are in fact friendly. The operator may therefore select other unidentified craft for pattern matching. In response to an operator selecting an unidentified craft track, the path monitor 810 provides a sequence of craft positions for the unidentified craft track to the pattern matcher controller 820. The pattern matcher controller 820 can select a pattern matcher 852a from the available pattern matchers 852a-n to perform the pattern matching. For example, an idle pattern matcher may be chosen. Alternately, where no pattern matchers are idle, the pattern matcher controller may be adapted to select the pattern matcher next likely to become idle, finish current processing. For example, the next pattern matcher to become idle may be based on processing state which can indicate what stage of the pattern matching process in currently being executed. Alternatively, a first in first out (FIFO) assumption may be applied by the pattern matcher controller 820 such that the pattern matcher that has been processing a pattern matching task the longest is assumed to be the next to finish, so a new pattern matching task will be scheduled for this pattern matcher.

In an alternative embodiment tasks, such as pattern matching and pattern mining, are broken down into a plurality of smaller jobs. For example, a pattern matching process may be broken down into a plurality of individual jobs such as:

searching the exemplar path database, dividing a tracked path into equidistant segments, dividing an exemplar path into equidistant segments, comparing distance and heading difference for each segment, segment based match probability

calculation, summing match probability etc. This plurality of jobs can be placed in a queue. The plurality of jobs may be created and place in the queue in response to an operator request for pattern matching. Any of the plurality of operators may request pattern matching. Each operator's plurality of jobs can be placed in the queue in order, but the jobs originating from different operators may be interspersed in the queue. For example, the queue may be implemented using a database or other data structure in a controller 820.

Similarly, pattern mining processes may be divided into a plurality of jobs and the pattern mining jobs placed in the queue interspersed with pattern matching jobs. A limit may be placed on the number of pattern mining jobs allowed in the queue, an alternative queue may be provided, or queue prioritisation may be applied to avoid pattern mining occupying too much of the system's data processing capacity and potentially delaying pattern matching. Alternatively, pattern mining may be restricted to times of low traffic and hence low pattern matching requirements.

In an embodiment a first in first out job queuing management regime is applied. However, other regimes may also be applied.

Each of the pattern matcher computers 825a-n may be executing a software application adapted to selectively perform any pattern matching or pattern mining job. The software applications can cause each computer 825a-n to request jobs from the queue. For example, the computer 825a may be adapted to periodically poll the controller 820 to request a new job if the computer 825a is not currently processing a job. All the data required by a computer 825a-n to complete a job can be stored in the queue. A job is claimed from the queue buy the computer 285a. Data regarding the computer 285a allocated the job and start time for the job may be recorded by the controller 820. Having this data recorded can prevent the job from being claimed from the queue by another computer 825b-n. The controller can also monitor the time taken for processing jobs in the queue. Where a job has been claimed for longer than a given time and no result returned, the controller may assume that a problem has occurred to cause a processing delay or hanging in the computer claiming the job. In this circumstance the start time and claimed computer for the job may be cleared so another computer can claim the job from the queue.

The computer 285a performs the processing for the job and returns the result to the controller 820. The result and end time for the job can be recorded. Before storing a result a check can be made of whether the computer retuning the result for the job is the computer claiming the job. Where there is a mismatch the result can be discarded to avoid corruption where a job was subsequently allocated to another computer. Once a job is completed, the computer 825a then pulls a further job from the queue. Similarly each of the other computers 825b-n are simultaneously extracting jobs from the queue, processing the job, and returning the result.

Where jobs are dependent on earlier jobs being completed the controller may be programmed to "hold" the job in the queue until all dependent jobs are completed. For example summing of match probability calculations may be dependent on a plurality of individual jobs to calculate match probability for each segment. As each segment probability result is returned these may be stored in a designated data structure associated with the summing job, only when this data structure is full may the job be claimed by one of the computers 825a-n for execution.

It should be appreciated that this distributed architecture has an advantage of being easily scalable. Further, as each computer operates independently and pulls new jobs form the queue, this provides an inherent redundancy in the system. For example, where one computer fails, even mid job, the load can be easily taken over by the other processing resources.

An advantage of using distributed pattern matcher architecture as illustrated in Figure 8 is that pattern matching can be performed simultaneously for a plurality of craft tracks. This can reduce the time required to characterise unidentified craft in the airspace. Further, it should be appreciated that the architecture is scalable by adding further pattern matchers. For example, doubling the number of pattern matchers can double the number of tracks that can be matched simultaneously. This can be particularly advantageous in a battle management context where rapid unidentified craft characterisation is critical.

The pattern matcher 525a performs pattern matching for the sequence of craft positions as described above with reference to Figure 2. The ranked list of exemplar paths is returned to the operator and displayed on the operator interface. Optionally the exemplar path track for each candidate path can be displayed. In some

embodiments the operator can selectively display each exemplar path. The operator may select the exemplar path, out of the candidate exemplar paths that the operator believes is the closest match. A further processing step can them be performed to determine using time information to correlate the selected exemplar path with an actual flight plan or flight number. For example, time of the craft position samples capture can be used to search data, stored in the exemplar path database, of the cluster of paths used to characterise the exemplar path. For example, this search may distinguish the tracked craft as a 9am flight from Melbourne to Adelaide from the generic Melbourne to Adelaide route.

In the embodiment illustrated in Figure 8, distributed processor architecture is shown for the pattern matcher functionality. However, distributed processing can also be applied for pattern mining. For example, the pattern miner functionality may be distributed across a plurality of processors such that each processor performs exemplar path characterisation as described with reference to Figures 4 to 6.

Depending on the embodiment, pattern mining may be distributed between individual pattern miners based on airspace regions or characterisation of sequences or clusters. For example, a single pattern miner or pattern miner controller may be adapted to identify common sequences of movement between regions. The characterisation of exemplar paths for each sequence, or cluster within a sequence, may then be distributed to different pattern miners. Adding further pattern miners can increase the amount of processing that can be performed simultaneously, hence improving the rate at which exemplar paths can be characterised for the airspace.

In an embodiment, distributed processors may be adapted to perform both pattern matching and pattern mining functionality. For example, the pattern matchers 825a-n may be implemented using a plurality of generic networked computers connected via a data communication network to a server in which the pattern matcher controller 820 and pattern miner 810 functionality is implemented. Each of the computers may be programmed with software modules for implementing pattern matching functionality and pattern mining functionality. The pattern matcher controller 820 and pattern miner 810 may be adapted to share the networked computer resources. For example, seven networked computers, each adapted to implement both pattern matching and pattern mining functionality, are provided. Where only three of these computers are required to meet the pattern matching demands at a given time, the remaining four computers may be utilised by the pattern miner. For example, the pattern miner may be adapted to periodically re-characterise regions of airspace based on recent historical data. The pattern miner may utilise spare capacity within the system for this purpose.

Where real time live craft pattern matching requirements increase, the pattern matcher may interrupt and take over processing resources being used by the pattern miner. For example, where the pattern matcher requirement increases from the capacity of three computers to five computers, two computers being utilised for pattern mining may be interrupted by the pattern matcher controller 820. Once the processing capacity required by the pattern matching decreases, for example from five to four computers, one of the previously interrupted computers can be released by the pattern matcher controller and again utilised by the pattern miner. As pattern mining is not time critical, the interruption has minimal impact of the pattern mining outcome. The interrupted computer may resume pattern mining from the point of processing where the interrupt occurred. Alternatively, the pattern miner may be adapted to return an error when a pattern mining process is interrupted. This enables the process to be restarted again using an alternative computer or by the same computer once it is released from pattern matching processing. Sharing resources in this manner has an advantage of increasing resource utilisation. Embodiments of the system described are adapted to perform exemplar path matching for tracked craft and pattern mining based on identification of common patterns between paths. Further embodiments are also adapted to perform

identification of patterns such as convergence or flocking. Algorithms for identification of convergence and flocking are described in articles: Johchim Gundmumdsson, Marc van Kreveld and Bettina Spekmann, "Efficient Detection of Motion Patterns in Spatio- Temporal Data Sets," Institute of information and computing science, Utrecht university, technical report UU-CS-2005-044, and Patrick Laube "Spatio-Temporal Data Mining - Coping with the Increasing Availability if Motion Data in Geography," SIRC, 2005.

Convergence patterns are patterns where paths intersect. For example convergence mining aims to identify two or more paths converging on a position at a similar time. For example, such a convergence pattern may indicate a rendezvous or provide warning of a potential collision. Convergence mining can search for arbitrary intersection events or intersections around a particular point of interest. A point of interest may be stationary, for example a building, or moving, for example the US president's plane Airforce 1 . A point of interest convergence test checks whether a predicted path passes within an area around the point of interest. For example, if the path passes within a radius of one kilometre of the point of interest. If an intersection is predicted the time of intersection can be calculated based on current tracked craft velocity. Where two or more intersections are predicted the time of intersection can be predicted and compared for each path. If the intersection is predicted to occur at a similar time for each path then a convergent event can be recorded and an operated alerted to the event.

The convergence event can be predicted based on current heading for the tracked craft or matched exemplar paths for the tracked craft. The live tracked craft data can be periodically tested for convergence, for example every minute. Alerting an operator to a potential convergence event can provide time for decision making.

Utilising matched exemplar path data for performing convergence mining may enable earlier prediction of convergence events. For example, the point of interest may be Airforce 1 . Based on a current commercial aircraft heading no intersection event would be detected. However, based on a matched exemplar path it may be predicted that a commercial aircraft will execute a change in direction which will cause the aircraft to pass near Airforce 1 . Using the exemplar path data enables the potential convergent to be predicted before the commercial aircraft executes the turn. This allows additional time for the operators to make a decision and provide guidance to the respective aircraft pilots. For arbitrary convergence a point of interest is not defined. In this convergence test an area is searched for predicted convergence events. For example, this may be useful for detecting a rendezvous before it occurs. For example, two ships converging at a point to exchange drugs or an aircraft approaching an aircraft carrier for landing. For arbitrary convergence an area is divided into a grid of cells, and a test for convergence performed for each cell. Tracks that are predicted to arrive at the same time generate a convergence event. Again, using exemplar paths may enable a convergence event to be predicted earlier than if only current headings are used. For example, in a remote area a regular mail delivery by light aircraft may follow a known exemplar path. A sight seeing plane may be identified in the same region, the sight seeing plane may or may not also be following an exemplar path. The exemplar path data may enable a convergence event, potentially a collision between the two aircraft, to be identified earlier than would be possible if only heading data was used.

Both the arbitrary and point of interest convergence algorithms can be adapted to enable an operator to specify the number of paths required to be converging before generating the event alarm. For example, for indentifying a rendezvous the number of paths will typically be two. However, in a battle situation convergences of more tracked craft may be of interests, for example convergence of five tracked aircraft may indicate forming up before an attack.

Real time craft tracking data may be used to identify flocking of a group of craft.

A flocking event occurs where two or more craft converge and change direction to follow a similar path. For example, flocking is similar to convergence detection or an extension of convergence detection described above where aircraft converge to fly in formation. Flocking event detection may also identify a flock leader. The flock leader is the craft following the path that others of the flock converge to. Exemplar path matching may be useful to distinguish unexpected flocking events from typical flocking events. For example a typical flocking event may be ships converging to follow a shipping channel through shallow water or aircraft approaching a runway of a land based airport. An unexpected flocking event may be aircraft taking off and landing from an aircraft carrier.

Embodiments of the present methods and systems can have significant advantages in reducing the load on operators involved in producing and maintaining recognised air pictures. Matching of tracked aircraft and vessels to exemplar paths can be performed using relatively short track segments. Further, potential exemplar path match results can be rapidly identified and presented to operators for decision making. The system can also perform periodic checking to identify an error in exemplar path matching or deviation of a matched path. This in turn can remove this responsibility from an operator.

Characterising actual tracked movement paths based on actual movement paths may enable paths to be matched with more certainty that matched based on theoretical flight plans. For example, using actual flight data can enable exemplar paths to include typical approaches to airports which may not ordinarily be available through regular planned flight path data. This can enable rapid and accurate path matching.

In the claims which follow and in the preceding description of the invention, except where the context requires otherwise due to express language or necessary implication, the word "comprise" or variations such as "comprises" or "comprising" is used in an inclusive sense, i.e. to specify the presence of the stated features but not to preclude the presence or addition of further features in various embodiments of the invention.

It is to be understood that, if any prior art publication is referred to herein, such reference does not constitute an admission that the publication forms a part of the common general knowledge in the art, in Australia or any other country.

Claims

1. A method of predicting a craft movement path comprising the steps of:

obtaining one or more craft position data samples for the craft;

determining a path match likelihood for each candidate exemplar path.

2. A method as claimed in claim 1 wherein the step of identifying candidate exemplar paths comprises the steps of:

selecting a first craft position from the craft position data;

for each identified proximate exemplar path:

3. A method as claimed in claim 2 wherein the step of determining relative heading between the exemplar path and craft movement path comprises:

selecting at least one further craft position from the craft position data;

4. A method as claimed in claim 2 wherein the step of determining relative heading between the exemplar path and craft movement path comprises determining heading difference between an absolute heading for the craft movement path obtained from the craft movement path data at the first position and an absolute heading for the exemplar path at a corresponding position.

5. A method as claimed in claim 2 wherein exemplar paths which pass within a given distance of the first craft position are identified as proximate the first craft position.

6. A method as claimed in claim 5 wherein the given distance can vary between exemplar paths.

7. A method as claimed in claim 6 wherein the given distance is mathematically defined for each exemplar path.

8. A method as claimed in claim 7 wherein the given distance is a function of a standard deviation for the exemplar path.

9. A method as claimed in claim 8 wherein the standard deviation is determined from known craft movements used for characterisation of the exemplar path.

10. A method as claimed in claim 2 wherein candidate criteria for selection of a candidate exemplar path include but are not limited to one or more of relative heading and relative path distance between the exemplar path and craft movement path.

1 1. A method as claimed in claim 10 wherein an exemplar path is excluded where the heading difference between the exemplar path and craft path is greater than a given threshold angle.

12. A method as claimed in claim 2 wherein the number of candidate exemplar paths is limited to a given number n.

13. A method as claimed in claim 12 wherein the n candidate exemplar paths having closest compliance with candidate criteria are selected.

14. A method as claimed in claim 1 wherein the step of determining a path match likelihood for each candidate exemplar path includes the steps of calculating a probability of match between the craft movement path and each candidate exemplar path, and ranking the candidate exemplar paths based on relative probability.

15. A method as claimed in claim 1 wherein the method is performed using real time captured craft position data for a detected craft.

16. A method as claimed in claim 1 further comprising the step of presenting the path match likelihood for a given number of exemplar paths to an operator for assisting the operator to characterise or identify the nature of the craft based on the craft's movement path.

17. A method as claimed in claim 1 further comprising the steps of periodically obtaining further craft position data, comparing with a chosen exemplar path and alerting an operator where deviation from the chosen exemplar path is identified.

18. A method as claimed in claim 1 further comprising the step of providing the repository of exemplar paths.

19. A method as claimed in claim 18 wherein providing the repository of exemplar paths comprises the steps of:

analysing historical data representing a plurality of tacked craft paths;

identifying common sequences of movement between regions;

identifying clusters of movement paths for the sequence; and

for each identified cluster:

storing the exemplar path in the repository.

20. A method as claimed in claim 19 wherein analysing a cluster to characterise an exemplar path comprises the steps of:

21. A method as claimed in claim 20 wherein the characterising craft position and distribution are determined based on root mean square analysis of the distance between each movement path.

22. A method as claimed in claim 1 wherein the exemplar path repository is a database storing exemplar path data searchable using geospatial database queries.

23. A craft movement prediction aid system comprising:

24. A system as claimed in claim 23 wherein the pattern matcher selects a first craft position from the craft position data for use to search the repository of exemplar paths based on the first craft position to identify any exemplar paths proximate the first craft position, and

for each identified proximate exemplar path:

determines relative heading and relative path distance between the exemplar path and craft movement path;

and

25. A system as claimed in claim 24 wherein the relative heading between the exemplar path and craft movement path is determined by selecting at least one further craft position from the craft position data, determining the distance between each further craft position and a position of the exemplar path which corresponds to the further craft position sample, and determining the relative heading based on difference in distance between exemplar path and craft movement path at the first vessel position and further vessel position.

26. A system as claimed in claim 24 wherein the relative heading between the exemplar path and craft movement path is determined by comparing an absolute heading for the craft movement path obtained from the craft movement path data at the first position and an absolute heading for the exemplar path at a corresponding position.

27. A system as claimed in claim 24 wherein exemplar paths which pass within a given distance of the first craft position are identified as proximate the first craft position.

28. A system as claimed in claim 27 wherein the given distance can vary between exemplar paths.

29. A system as claimed in claim 28 wherein the given distance is mathematically defined for each exemplar path.

30. A system as claimed in claim 29 wherein the given distance is a function of a standard deviation for the exemplar path.

31. A system as claimed in claim 30 wherein the standard deviation is determined from known craft movements used for characterisation of the exemplar path.

32. A system as claimed in claim 31 wherein candidate criteria for selection of a candidate exemplar path include but are not limited to one or more of relative heading and relative path distance between the exemplar path and craft movement path.

33. A system as claimed in claim 32 wherein an exemplar path is excluded where the heading difference between the exemplar path and craft path is greater than a given threshold angle.

34. A system as claimed in claim 24 wherein the number of candidate exemplar paths is limited to a given number n.

35. A system as claimed in claim 34 wherein the n candidate exemplar paths having closest compliance with candidate criteria are selected.

36. A system as claimed in claim 23 wherein determining the path match likelihood for each candidate exemplar path includes ranking relative probability of match between the craft movement path and each candidate exemplar path.

37. A system as claimed in claim 23 wherein the pattern matcher uses real time captured craft position data samples for a detected craft.

38. A system as claimed in claim 23 further adapted to output the path match likelihood for a given number of exemplar paths to an operator for assisting the operator to characterise or identify the nature of the craft based on the craft's movement path.

39. A system as claimed in claim 23 further comprising a pattern miner adapted to characterise exemplar paths from historical craft movement data and populate the exemplar path repository with exemplar path data.

40. A system as claimed in claim 39 wherein the pattern miner analyses historical data representing a plurality of tacked craft paths, identifies common sequences of movement between regions, identifies clusters of movement paths for each sequence, and for each identified cluster analyses the cluster to determine a characterising exemplar path for the cluster comprising a characterising craft movement path and craft distribution about the characterising craft movement path for the cluster.

41. A system as claimed in claim 40 wherein the pattern miner analyses a cluster to characterise an exemplar path by:

42. A system as claimed in claim 41 wherein the characterising craft position and distribution are determined based on root mean square analysis of the distance between each movement path.

43. A system as claimed in claim 23 further comprising an exemplar path repository in the form of a database storing exemplar path data searchable using geospatial database queries.

44. A system as claimed in claim 23 comprising a pattern matcher controller and two or more pattern matchers in data communication with the pattern matcher controller and each pattern matcher executable using separate processing hardware resources and wherein the pattern matcher controller is adapted to allocate one or more sequences of craft position data samples to each pattern matcher for

identification of candidate exemplar paths.

45. A pattern matcher adapted to identify, based on one or more craft position data samples, any one or more candidate exemplar paths from a repository of exemplar paths each representative of a previously characterised craft movement path by:

determining a path match likelihood for each candidate exemplar path.

46. A pattern matcher as claimed in claim 45 wherein the relative heading between an exemplar path and craft movement path is determined by selecting at least one further craft position from the craft position data, determining the distance between each further craft position and a position of the exemplar path which corresponds to the further craft position sample, and determining the relative heading based on difference in distance between exemplar path and craft movement path at the first vessel position and further vessel position.

47. A pattern matcher as claimed in claim 45 wherein the relative heading between the exemplar path and craft movement path is determined by comparing an absolute heading for the craft movement path obtained from the craft movement path data at the first position and an absolute heading for the exemplar path at a corresponding position.

48. A pattern matcher as claimed in claim 45 implemented using a plurality of distributed data processing resources.

49. A pattern miner adapted to characterise exemplar paths from historical craft movement data by:

analysing historical data representing a plurality of tacked craft paths;

identifying common sequences of movement between regions;

identifying clusters of movement paths for each sequence;

analysing each cluster to determine a characterising exemplar path for the cluster comprising a characterising craft movement path and craft distribution about the characterising craft movement path for the cluster; and populating an exemplar path repository with exemplar path data.

50. A pattern miner as claimed in claim 49 wherein a cluster is analysed to characterise an exemplar path by:

51. A pattern miner as claimed in claim 49 implemented using a plurality of distributed data processing resources.

52. Computer program code which when executed causes a computer to implement a craft motion path prediction method as claimed in claim 1.