K.Pack. Automated false track identication in Proc

Automated False Track IdentificationKenneth PackScience Applications International Corporation 400 Virginia Avenue SW, Suite 800Washington DC 20024Abstract1Suboptimized real-time surveillance systems trade off reporting all targets in the surveillance volume against minimizing the number of false tracks. Analysis of archived data can identify false tracks. By automating the analysis, system optimization can be done sooner. Automated analysis identifies candidate false tracks using track characteristics over the life of the track. These characteristics depend on the sensor data and operational environment. This paper discusses automated false track identification of ASDE-X (Airport Surface Detection Equipment Model X ) tracks collected at multiple airports. These tracks are supported by radar and multilateration sensor data.IntroductionMany real-time systems, including ASDE-X, fuse multiple detections from one or more sensors to form target tracks. Sometimes, ASDE-X displays a non-existent, or false target. When a vehicle or aircraft transmits a Mode-A or ICAO2 identifier, ASDE-X uses the identifier to tag the target. For this paper, the false track definition is extended to include tracks having a wrong position3, and tagged tracks having a wrong identifier.The ASDE-X system1 has one or more surface movement radars (SMR) and a multilateration subsystem. At airports where an ASDE-3 radar is available, ASDE-X uses the ASDE-3 radar instead of installing a new SMR radar. ASDE-X integrates data from the airport surveillance radar (ASR) with its own sensor data to detect, identify 1 This work was partially supported by the Federal Aviation Administration under FAA Contract DTFA01-01-0-0021.2 ICAO: International Civil Aviation Organization. Mode-S transponders transmit ICAO identifiers.3 For example, a plane with its transponder on might generate a multipath track on a taxiway.and track aircraft and ground vehicles in the surveillance volume. The multilateration subsystem uses signals from aircraft Mode-A and Mode-S transponders to identify and locate aircraft and transponder equipped vehicles. The multilateration subsystem locates the transponders, using the time difference of arrival of the transponder's signal received at multiple sensors. The ASDE-X surveillance volume includes the airport taxiways, runways, and approach corridors.In ASDE-X, target tracks can be either displayed or not displayed. The system is configured to not display targets in the gate area. For false track analysis, the gate area is considered outside the surveillance volume. Undisplayed tracks and sensor position estimates, called plots, are available for analysis throughout the surveillance volume and the gate area. All displayed tracks are updated once a second. If there are no new supporting plots, the track's position is updated based on its previous location, course, and speed.During system optimization, analysts adjust configuration parameters to minimize the number of false targets without losing real targets. The optimization process uses archived data collected 24 hours a day, seven days a week for several months. With archived data, the entire track history of each target can be reviewed multiple times by several analysts. Details of track initiation, movement, track features, and track death are kept in the archives. Since truth about the real targets present is not available, differentiating between real and false targets can be a time consuming process. Rule Development for Identifying False TracksFor targets of opportunity, false track identification is based on events or features that deviate significantly from those of a typical real target. Fortunately, at airports, airplane and ground1vehicle tracks have predictable events and identifiable features. These targets enter and leave the surveillance volume along its edge. Tracks of real targets tend to be contiguous. During their track life, aircraft typically travel on a runway. On the runway, they either decelerate or accelerate to a relatively high speed. Landing and departing aircraft have surface radar, ASR radar, and multilateration support during their track life. Ground vehicles travel along the taxiways, runways and access roads in the airport movement area. Parked vehicles in the movement area eventually move when maintenance personnel have finished their work.At a mid-sized airport, during the busiest hours of a day, the average number of aircraft arriving can be 10 to 20 per hour. At a large busy airport, the average number of aircraft arriving can be more than one per minute for over 10 hours. On average, the same number of aircraft leave as arrive. In addition to aircraft, maintenance vehicles travel throughout the movement area.Because it takes time to identify false tracks, analysts developed rules to improve their productivity. These rules evolved over time. The relative usefulness of the rules changed as ASDE-X changed. For example, searching for tracks that existed a long time without sensor support was once a good rule for finding false tracks. However, the default configuration now terminates unsupported tracks more quickly.When a new rule is coded into either a MATLAB script or a compiled program, it is evaluated using a binary hypothesis test. The two possible outcomes are:(1)The rule identifies a track as false whenthere is a consensus among analysts thetrack is false.(2)The rule identifies a track as false whenthere is a consensus among analysts thetrack is real.Since a false track can theoretically be missed, a set of tracks is sometimes manually searched for false tracks missed by all the coded rules.Often a false track identification rule classifies a track as false when it is real. Some rules have fewer misclassifications than others. Todifferentiate between rules with better prediction rates than those more likely to misclassify a real target, the program assigns a low or high confidence to each false track prediction. When an analyst has time, all the predicted false tracks can be evaluated; otherwise, only the high confidence predictions are evaluated.The features used by the false track identification algorithm can be categorized into:(1)track life,(2)track proximity,(3)sensor specific features, and(4)track dynamics.Track life features include how long the track existed, how long the track was displayed, and the time between when the track was initiated and when it was first displayed.Track proximity features include the number of tracks that were close to the track of interest, and whether the track initiated or terminated close to another track. Track proximity features for pairs of tracks that were once close to each include the number of times the two tracks were near each other. Rules based on track proximity find radar splits and duals. A radar split occurs when an aircraft is detected by the surface radar as two or more plots, and ASDE-X creates a track for each plot. A dual occurs when two sensors, such as multilateration and a radar each detect the same target, but ASDE-X creates a separate track for each sensor.Sensor-specific features include identifying which sensors supported a track, and when each sensor first supported that track. The ASR and multilateration systems associate a Mode-A code with the tracks they report. These codes tend to be unique, so two tracks should not simultaneously have the same code4. The multilateration system associates ICAO codes with Mode-S equipped targets. These codes are unique. The surface radar provides target extent. The larger a target is in range and azimuth, the less likely it is to be a multipath false track.4 There are cases where multiple transponders are either assigned a non-discrete code or the pilot has entered the wrong Mode-A code in the transponder.2Track dynamic features include the distance the track traveled over its lifetime, where the track started, where it ended, whether it was ever in a near runway region, whether it was ever on a runway, and whether it ever traveled fast. If the track was on a runway, the speed at which the track entered and exited the runway are included in the dynamic feature set.As with the rules, the features were identified by analysts searching for clues as to why a track should or should not be classified as false. In general, the more relevant features identified, the more likely a track is correctly classified by the automated false track identification program.During rule development, analysts identified rules that are more relevant in one region of the airport than in another. Currently, the rules use three regions: called runway, near runway, and nuisance. Each region is made up of one or more polygons. The polygons do not have to be connected. For example, the runway region covers each runway. The near runway region typically covers each runway and extends 500 feet from the runway edges. This region can be modified to adjust the sensitivity of the rules that depend on it. For example, many rules expect real targets, while in a near runway region, to have tracks that are updated each cycle with supporting plots. The third region is the nuisance region. This region is used to identify construction zones, where barricades and construction equipment make it difficult to identify false tracks.It is possible the automated false track program will miss one or more false tracks. The analyst must choose between reviewing every detail of the archived data for a limited time period, or to examine as much data as the analyst's schedule permits. It takes time for all combinations favorable to false track formation to manifest themselves at an airport. False tracks formation depends on the relative geometry of aircraft, vehicles, fixed structures, radar, and multilateration sensor units. The traffic flow in one direction sets up false track conditions different from those in the opposite direction. Multipath conditions can depend on large aircraft parking at a particular gate. Whereas weak transponders are difficult to detect, powerful ones at the same location can generate multilateration false plots. Some aircraft have only a bottom mountedantenna connected to their transponder. Multipath MLAT tracks tend to form more often on signals from these aircraft than with aircraft having top mounted antennas or dual antennas. Some transponders possibly do not meet their specifications2. False track formation also depends on rain, snow, and sometimes, grass that needs to be mowed.RulesAs previously mentioned, candidates identified by some rules are more likely to be false tracks than those predicted by other rules. The purpose of false track analysis is to improve ASDE-X's performance. False track analysis was automated so more data could be analyzed, reducing the time to optimize each new installation and evaluate upgrades at existing sites.The composite rule set is designed to find as many false tracks as possible while minimizing the number of real targets incorrectly tagged false. Due to the interaction between analysis and development, lessons learned from false track analysis generate ASDE-X software and adaptation changes. It is expected that many of the current high confidence rules used in the automated false track identification software will be modified or replaced in the future. Historically, the most likely modification is to add more tests to each rule.The prediction rules are listed below. Each rule is summarized by the type of false track it is designed to identify.(1) Static SMR MPThis rule is designed to identify false tracks due to static multipath. Multipath can be either static or dynamic. When the multipath track remains stationary, it is static. If the multipath track moves, it is dynamic. Static multipath can be due to fixed structures or parked aircraft in the gate area. This problem is worse when an airport terminal is between the SMR radar and the airport movement area, than for other geometries. The rule also finds false tracks due to snow banks, pooled rain water on the movement area, and other non-vehicle reflectors. In the spring and summer, it has found false tracks from tall grass growing along a taxiway.3A track is predicted to be static SMR MP if:•it has SMR support,•it is a small target,•it ends near where it starts, and•its average speed is less than a slow speed threshold.If the track has SMR support, ends near where it starts, and its average speed is below the slow mover threshold, but it is not a small target, then the rule predicts the track is false, but assigns a low confidence to the prediction.Alternatively, if three or more tracks "pass through" a track, the program predicts the passed through track is a multipath false track. The pass through criteria is that the tracks come within 80 feet of each other.(2) Dynamic SMR MPThis rule is designed to identify false tracks created by dynamic SMR multipath. Dynamic multipath occurs when at least one of the reflectors moves. Dynamic multipath tracks often occur when either the primary reflector is in a region masked out by the radar processor or the system has underestimated the extent of the primary reflector. The primary reflector is a reflector between the radar and the false track. In either case, the operational system fails to identify the false track as multipath.A track is predicted to be dynamic SMR MP if:•it has SMR support•it is a small target•it is unsupported for multiple consecutive track updates,•at least one of the unsupported updates is ina near runway region, and•it is either unsupported for most of its life or its life is short.This rule illustrates that not all false tracks are treated the same. Those closer to or on a runway are more serious than those far from a runway. This rule misses dynamic SMR multipath tracks occurring outside the near runway region. Depending on time constraints, the analyst can expand or contract the near runway region to coverall taxiways or concentrate on the movement area closer to the runways.(3) MLAT MP CriteriaThis rule is designed to identify false tracks created when the position estimate is corrupted by multipath signals at one or more of the multilateration receivers.A track is predicted to be MLAT multipath if:•it has had MLAT support,•it does not have a long period withconsecutive MLAT supported updates, •after the first MLAT supported update, less than a third of the track updates haveMLAT support, and•for much of its life, the track is unsupported by any sensor.(4) Duplicate IDThis rule identifies when two tracks have the same aircraft identifier (ACID). Duplicate identifiers occur when one track steals the ACID of another.The rule is invoked when either:•two tracks have the same ACID at the same time, or•one track ends within a couple seconds of the second track's start, and•both tracks have the same ACID.(5) Multiple ACIDThis rule identifies when a track has more than one ACID during its lifetime. Multiple ACIDs occur when a track misassociates with the multilateration or ASR support for another target.(6) Arriver DualThis rule is designed to identify handoff failures between the ASR and surface radar for an arriving aircraft.This rule has three parts:•Unsupported Arriver,•Track Initiated on Runway, and•Pairing4A track is tagged an unsupported arriver if either•it enters a runway region,•it has multiple ASR supported updates,•it travels over 10,000 feet,•it is tracked for at least a minute,•its last support is from an ASR, and•its last update is unsupported by anysensors,or•it has multiple ASR supported updates,•it travels over 10,000 feet,•it is tracked for at least a minute, and•its average speed is above a high speed threshold.A second track is tagged as initiated on a runway if•its initial report is on a runway,•it lives long enough, and• a sufficient percentage of its updates have sensor support.For this rule, the minimum time an initiated on runway track needs to live is 45 seconds. The longer a target lives, the more unsupported updates it may have and still be used in the pairing rule.The second track is identified as a dual and paired with the first if•both tracks are ever in a near runwayregion,•the second track starts within 30 seconds after the first track enters a near runwayregion, and•the first track, when it entered the nearrunway region, is near where the secondtrack starts.(7) Coasted DepartureThough this rule sometimes finds duals formed on departure, it more often finds aircraft departing without their transponder on and small aircraft that bank too sharply for the tracker to associate new reports with the existing target track.A track is flagged as having a coasted departure if•it travels a long distance in its lifetime,•it lives over a minute,•it has few or no plots with ASR support,•it dies without sensor support, that is itcoasted,•it leaves a runway at a high speed, and•it has multiple updates when it is travelling above a high speed threshold.(8) ASR DualThis rule is designed to detect an ASR/MLAT dual when an airplane is flying.Two tracks are flagged as duals when at least one of the tracks has 25 or more updates where •both tracks are travelling at a high speed,•neither track has an invalid Mode-A code or ICAO code,•neither track has SMR support, and•the tracks are near each other.(9) ASR MisassociationThis rule is designed to detect when an ASR plot associates with the wrong track.A track is predicted to be false if•it initiates with non-ASR sensor support,•it has multiple updates before getting any ASR supported updates,•it has only a few ASR supported updates,•most of its updates are unsupported.(10) ASR MultipathThis rule is designed to detect when ASR multipath plots form a track.A track is predicted to be false if•it initiates with ASR support•it has only a few ASR supported updates, and•most of its updates are unsupported.(11) MLAT DualThis rule is designed to detect SMR/MLAT duals. This rule first searches for pairs of tracks that were close to each other. If the pair were close to5each for a significant number of updates, then the pair is flagged as potential split track pair.The pair is flagged as a dual if•one of the tracks has multiple unsupported updates and its last update is unsupported, •one of the tracks starts within 300 feet of where the two tracks were first close toeach other, and•after the tracks have been within 100 feet of each other for at least 20 updates, they aremore often within 100 feet of each otherthan apart by more than 100 feet,•one of the tracks has multilaterationsupport, and•the first MLAT supported update occurs within five seconds of the first time thetracks are close to each other.(12) Radar SplitThis rule is designed to detect false targets due to the radar reporting multiple plots for an aircraft. The rule first searches pairs of tracks. If either track has many updates close to the other, the pair becomes a potential split track pair.The pair is flagged as a split if•one of the tracks has multiple unsupported updates and its last update is unsupported, •one of the tracks starts within 300 feet of where the two tracks were first close toeach other, and•after the tracks have been within 100 feet of each other for at least 20 updates, they aremore often within 100 feet of each otherthan apart by more than 100 feet, and •neither track has MLAT support.(13) MisassociationThis rule identifies when a sensor plot from one target associates with the track of another target and causes the second track's position to be wrong.The rule first searches pairs of tracks. If either track has many updates close to the other, the pair becomes a potential split track pair.The rule then checks each potential split using the following three tests. If the pair meets all thecriteria of any one test, the pair is flagged as a misassociation.(1) The pair is flagged as a misassociation if•one of the tracks has multiple unsupported updates and its last update is unsupported, •one of the tracks starts within 300 feet of where the two tracks were first close toeach other, and•after the tracks have been within 100 feet of each other for at least 20 updates, they aremore often within 100 feet of each otherthan apart by more than 100 feet.(2) The pair is flagged as a misassociation if•one of the tracks has multiple unsupported updates and its last update is unsupported,and•the two tracks are close to each other forover a hundred updates.(3) The pair is flagged as a misassociation if•either track is a landing or departingaircraft.(14) ClutterIf a track is in multiple potential split track pairs, it is categorized as clutter.(15) No SMR SupportThis rule predicts tracks that never had SMR support are false.(16) Initiated on RunwayThis rule is designed to detect tracks having non-normal behavior and not categorized by other rules. This rule predicts a track is false if it initiates on a runway. Sometimes tracks tagged via this rule are matched with arrivers having no support; together the pair form a dual.(17) Late DisplayThis rule is designed to detect tracks having non-normal behavior and not categorized by other rules. This rule predicts a track is false if the track is on a runway shortly after it started. The6effectiveness of this rule depends on how close the runways are to the terminal area, and the minimum time threshold that determines whether or not the track reached the runway too quickly.(18) Dropped TrackThis rule is designed to detect tracks having non-normal behavior and not categorized by other rules. These tracks are sometimes created by a misassociation, but were not caught by the misassociation logic.A track is predicted to be false if•it did not travel at a high speed•it lived over a minute•it traveled more than a minimum distance•it ended in a near runway region. SummaryThis paper presents an approach used to find false tracks in ASDE-X data collected during site optimization. The data consists of targets of opportunity and false tracks. Rules, used in an automated false track identification program, are provided. These rules were developed based on a profile of normal aircraft and vehicle tracks. The candidate tracks output by the program are reviewed by an analyst to verify whether or not they are false.AcknowledgementIt is a pleasure to acknowledge the hard work done by Hong Kaing building the software and regression testing the algorithms. I am also grateful to insights provided by George Go, cataloging the different types of radar false tracks and characteristics useful for identifying them. I wish to thank Karim Belamqaddam, Matias Palavecino, and Mark Scheufler for their evaluations of the tracks identified by the false track identification software.2007 ICNS Conference1-3 May 20071 National Airspace System (NAS) SubsystemLevel Specification for Airport Surface DetectionEquipment Model X (ASDE-X), FAA-E-2942Department of Transportation, Federal Aviation Administration, Version 1.1, August 13, 20012 Talotta, Nicholas J., et al, "A Field Study of Transponder Performance in General Aviation Aircraft",U.S. Department of Transportation, Federal Aviation Administration, DOT/FAA/CT-97, December 5, 19977。