Fusion of Data from AIS and Tracking Radar for the Needs of ECDIS Witold Kazimierski Maritime University of Szczecin Faculty of Navigation, Chair of Geoinformatics Szczecin, Poland w.kazimierski@am.szczecin.pl Andrzej Stateczny Maritime University of Szczecin Faculty of Navigation, Chair of Geoinformatics Szczecin, Poland a.stateczny@am.szczecin.pl Abstract The paper presents results of the research on problem of radar-ais data fusion in ECDIS. First the concept of these systems according to IMO is provided, then a set of theoretical information concerning radar tracking and the fusion itself is given and finally numerical research with real data are presented. Two methods of decentralized fusion are examined with the analysis of their parameters. A proposal of calculating covariance matrix for radar and AIS data is given. The research are followed with the conclusions. Keywords data fusion, navigation, ECDIS, target tracking, radar, AIS I. INTRODUCTION One of key aspects of modern maritime navigation is data integration. It is of essential meaning for the officer in charge of navigational watch to receive reliable and accurate complex information from various sensors available on board of the vessel. Various systems are being developed independently, however simultaneously new systems using integrated data are appearing. First example of such system is ECDIS (Electronic Chart Display and Information System) introduces at the end of 20 th century. The main idea was to present navigational information with the background of Electronic Navigational Chart (ENC). With time the system has been developed and new functionalities arose. Apart of others, functions, which are presenting collision situation around own ship have been evolving. The basis for this is fusion of data from tracking radar and AIS (Automatic Identification System) [1]. Tracking radar is the most important sensor used for so called tactical navigation. It is currently used on the most of merchant vessels, giving independent information about movement of targets around own ship. Main deficiencies of this solution are the delay of tracking, mostly during maneuvers and lack of identification [2]. An alternative is to use AIS. It is the system in which the vessels are transmitting dynamic and static data from own sensors via VHF. Thus, an accurate information about other ships, together with identification is transmitted. The biggest problem is that data comes external sensors and not from independent observation with own sensor. Each error of sensor in target ship results in improper data, which are transmitted on air. In practice both systems (tracking radar and AIS) are used simultaneously and the fusion of them has to be made. The most important platform for fusion is ECDIS, which is sometimes introduced into INS (Integrated Navigational System). Radar-AIS fusion is also a basis for navigational decision support systems like in [3], [4], [5]. The presents concept and problems of tracking radar AIS data fusion from ECDIS s point of view. First the ECDIS itself is described based on IMO requirements. Then the most popular concepts of fusion are given. Finally the numerical experiment is presented in which a comparison of two approaches and a proposal of modification of them are given. II. ECDIS ACCORDING TO IMO The newest requirements for ECDIS are given by IMO in Resolution MSC.232(82) adopted in 2006 [6]. According to this ECDIS is a navigation information system which displays selected information from a system of electronic navigational chart (SENC) with positional information from navigation sensors to assist the mariner in route planning and route monitoring, and if required display additional navigationrelated information. It is very important that with adequate back-up arrangements ECDIS can be accepted as complying with the up-to-date chart requirements of SOLAS. ECDIS may be implemented of board in either of two ways: dedicated standalone workstation; a multifunction workstation as part of an INS. A. Main functions The primary function of ECDIS is to contribute to safe navigation. More detailed functions are related mostly to navigational charts. ECDIS should: be capable of displaying all chart information necessary for safe and efficient navigation; facilitate simple and reliable updating of the ENC; reduce the navigational workload compared to using the paper chart;
provide appropriate alarms or indications with respect to the information displayed or malfunction of the equipment. Detailed operational and functional requirements for display of ENC, stated in the Resolution and also in IHO standards are beyond the scope of this paper. It is however important that requirements for displaying other navigational information are also given. These include radar and AIS data. B. Display of target data Although ECDIS is focused on presenting chart information, displaying of other navigational information for enhancing navigational safety is also allowed. The most important data included in the Resolution are radar and AIS. According to [6] they can be transferred from systems compliant with suitable IMO standards and added to the display. However they should not degrade SENC information and should be clearly distinguishable from it. The possibility of removing radar/ais data by single operator action if needed should be ensured. No further requirements are stated. Radar information transferred to ECDIS can include both, radar image and/or tracked target information. It is very important that added navigational information should use a common reference system with SENC. The radar image and the position from the position sensor should both be adjusted automatically for antenna offset from the conning position. It can be noticed while analyzing requirements for ECDIS, that their focus mostly on displaying chart. AIS and radar are only mentioned and the reference to other IMO documents like [7] or [8] is made. The concept of integration presented in [7] assumes in fact only target association and selection one of the targets radar or AIS. It can be seen that in fact no advanced fusion algorithms are needed to fulfill these requirements, except harmonized criteria for association. Nevertheless, presenting of integrated radar-ais information on screen of ECDIS is commonly used and plays an important role in modern navigation. III. AIS RADAR FUSION CONCEPT In general two major concept of AIS-radar fusion can be presented, namely decentralized and centralized approaches. In the first one, complex information is first calculated in each sensor and then provided to a fusion algorithm, where it is integrated with established rules. In the second one, raw measurements from sensor are transmitted for further processing to fusion module. In case of ECDIS it is more convenient to use decentralized concept. Radar and AIS are for ECDIS just an external source of data and creating additional filter in ECDIS for centralized fusion is pointless. Decentralized approach based on Kalman filter is also the most popular in literature and presented for example in [9] or [10]. The first step for all fusion algorithms is target association. It is the basis and very often this step generates most of the problems. Various algorithms for this task are also implemented including numerical calculations, grey theory or fuzzy logic. Proposed solution and a survey on association problems can be found e.g. in [11]. A. Problems of association Differences between tracking radar and AIS concepts causes that tracks received from both systems, although describing the same target, are of different nature. Therefore it is not a trivial problem to properly associate radar and AIS tracks. Among the main problems in association process following can be formulated [12]: lack of time synchronization between measurements in both systems; various time interval of measurements; different speeds and courses (dualism); lack of identification of radar target; large differences in position accuracy; size of radar echo in relation to point AIS target. Most of this problems can be solved with methods presented in literature, and their description is beyond the scope of paper. However it is worth of mentioning, that the values used for track fusion are already some kind of an approximation of data and not the real measurements. Many of the problems are caused by inaccuracies in radar target tracking, therefore the most popular methods are briefly presented below. B. Radar tracking methods The literature for radar tracking is relatively wide. A small fraction of it however concerns tracking in marine systems/ radars. Not all methods presented for air traffic, e.g. in [13] can be used in marine environments and the parameters of them needs to be usually adjusted. This is because of characteristics of movement of targets at sea. Marine radar trackers have to deal with relatively slow movement ahead comparing to large transversal errors. Various numerical filters were proposed for this task all over the years. The most popular of them is Kalman filter with its numerous modifications widely described in literature. It is however possible to use other approaches like artificial neural networks. They seems to be an interesting alternative to numerical filters and therefore are worth of mentioning. Artificial neural networks is the group of algorithms, that has non-linearity implemented in its nature. Therefore they should perform quite good in case of nonlinear movement. The research on this has been carried out in Maritime University of Szczecin for the last 15 years. Many network structures had been examined and especially good results have been obtained with the use of General Regression Neural Network [14]. The research [15], [16], [17] showed that due to considerable differences in dynamics, uniform rectilinear motion and non-linear motion require the application of different GRNN parameters. Thus a concept of multiplemodel neural filter arose [18]. Results of verification research presented in [19], [20] and [21] has shown that neural filter is a real competition for commercially used filters, especially during target maneuvers.
Apart from tracking method it is of essential meaning to use the most reliable positioning method for obtaining plots in radars. This task means mostly extraction of targets form the radar screen. Another interesting alternative can be shown here for traditionally used pulse radars, which are FMCW radar particularly good for inland waters and short distances due to their known advantages [22], [23], [24], [25]. The research carried out so far proved their usefulness for comparative navigation [26], [27], [28] and for spatial sensor planning [29]. The task of obtaining plots and target data in AIS is much simpler as it is an external sensor and there are not many possibilities of adjusting it. One can accept the data or not. C. Track fusion algorithms When the tracks are received and associated a process of track fusion begins. It is assumed that the state vector and covariance is known from both systems (radar and AIS) and that they describe the same target. Various algorithms for track fusion are presented in literature e.g. in [13], [30], [31]. From these the most popular seems to be simple fusion; with the use of cross-covariance. In the simple fusion algorithm presented e.g. in [9] or [30] the fusion is a weighted average of elementary estimates (x), where the weights are computed directly from covariance (P). For radar-ais tracking of one target the fusion equation has a form of x = (P -1 r +P -1 a ) -1 (P -1 r x r +P -1 a x a ), (1) with error covariance matrix P = (P -1 r +P -1 a ) -1. (2) The case of calculating fusion with cross covariance is more complicated and in a classical from it requires more information from elementary filters. According to [30] the fusion for two sensors can be calculated as follows: where x = x a + W(x r - x a ) (3) W = (P a P ar )U ar -1-1 U ar = P a + P r - P ar P ar (5) where P ar is the cross-covariance matrix, calculated recursively with the use of Kalman Filter s matrixes of elementary filters. Such a solution is useless if only estimate and its covariance is known and no more details about elementary filters. This is a situation like in ECDIS, where no information about tracking filter is transmitted, only the values. This situation also can happen if other method than Kalman filter is user, like above mentioned neural method. Thus, a method of approximation of cross-covariance matrix by the Hadamard product of input matrices was proposed in [13] and [30]: P ar = ρ(p a P r ) 1/2 (6) where ρ is an effective correlation coefficient, determined empirically. In these research value of 0.4 is taken following [30]. (4) In literature other algorithms for track association can be found, however these two seems to be most popular. IV. NUMERICAL EXPERIMENT A numerical experiment for researching radar AIS fusion, based on real data was proposed. The main goal was to compare simple fusion with cross-covariance fusion algorithm and to propose a values for covariance. ECDIS environment was assumed, i.e. only values of state vector received from external sensors (radar and AIS) are known. For track association a three-step algorithm, consisting of position association, track association and history correlation was performed as per [11]. Thus the targets were assumed to be associated and only track fusion was examined in this research. A. Research concept The research focused on analysis of methods and variance matrices. Three stages of the research were proposed: comparison of fusion algorithms; comparison of variation matrices; length of sliding window analysis. In the first stage the algorithms of simple fusion and of fusion with the cross-covariance were examined. The state vector was formulated as x = [BE, D, COG, SOG] T (7) where: BE bearing, D distance, COG course over ground, SOG speed over ground. After association process both vectors were synchronized and COG and SOG was known for both sensors. All the values in state vector were treated as independent measurements. Thus the variance matrices in the first stage had a form of diagonal matrix P = diag(σ 2 BE,σ 2 D,σ 2 COG,σ 2 SOG ) (8) For the radar, particular values were taken after IMO requirements P r = diag(4, 2500, 25, 0.25) (9) For AIS target, particular values were taken based on so called relative accuracy P a = diag(0.04, 225, 9, 0.0001). (10) In the second stage of the research only one selected fusion method was used, but the values of errors covariance matrices were changed. Modification of covariance matrices was proposed. The values in state vector was treated as samples of measurements of variable. Thus sample variances for the set of values over a sliding window were proposed as items in covariance matrices, as in (11): P v =diag(var(be k-l :BE k ), var(d k-l :D k ), var(cog k-l :COG k ), var(sog k-l :SOG k )) (11) where l is the length of sliding window. To remain the influence of sensors accuracy, the covariance matrix used in
this stage of research was a Hadamard product of (8) and (10), resulting in matrix (12). P=diag(σ 2* BE var(be k-l :BE k ),σ 2* D var(d k-l :D k ), σ 2* COG var(cog k-l :COG k ), σ 2* SOG var(sog k-l :SOG k )) (12) Analysis of the influence of the length of sliding window was carried out in the third stage of the research. This observation confirms that both fusion methods are performing task accordingly to assumptions. It can be also noticed that cross-covariance method relies more on the covariance matrix, as this fusion is closer to AIS data. B. Research scenario The research was performed in the own prepared software in VisualBasic. Net. The software allows to implement any fusion method and to adjust easily parameters of them. Data for the scenarios can be simulated, imported off-line from files or received on-line via serial ports. In the research the data were imported from previously recorded files. Data for the research were recorded on research-school ship NAWIGATOR XXI in the southern Baltic sea. NMEA strings from tracking radar and from AIS were recorded and then play backed in software. Scenario presented in the research in this paper included radar and AIS observation of general cargo with the LOA of 95 meters and DWT of 4 000 tons. 15 minutes of observation with more or less uniform motion is presented. Also own ship was not maneuvering. Trace of the target received on the radar screen is presented in Fig. 1. Fig. 2. Comparison of course estimated with different fusion methods B. Comparison of variances In the second stage of research only simple fusion method is used, but another covariance matrix, based on dynamic measurements of variance (called on the figure sample variance) is also used, alternatively to covariance matrix based on accuracies. The results are shown in Fig. 3. Fig. 1. Trace of radar target in the experiment, relative to own ship position V. RESEARCH RESULTS The results of the numerical experiment are presented for three research stages described above. As the basis for presenting results it was decided to present course in the function of time. A. Comparison of fusion algorithms In this stage the fusion with two methods simple fusion and cross-covariance - was performed. The variance matrix was stated based on IMO requirement. The comparison of course estimation is presented in Fig. 2. When analyzing Fig.2 it can be noticed that the estimated fusion is much closer to AIS data, which is understood, as these data are much more accurate. However both fusions deviates a bit towards radar course, following also its values. Fig. 3. Comparison of course estimated with different variances for simple fusion The graph called Accuracies is the same as in case of figure 2. Graph called Sample variance show the case where simple fusion is used but covariance matrix is calculated according to (12) and the length of sliding window is 10. It can be noticed that fusion in this case follows more stable sensor. At the beginning when radar values are varying a lot, fusion is almost equal to AIS, but when AIS data begin to vary, fusion deviates into radar. This interesting feature might be used for detecting temporary errors of any sensor. The problem of maneuvers might however occur. To make better analysis of this issue, another step of research is proposed.
C. Length of sliding window analysis In this stage simple fusion is used with covariance matrix based on sample variance, however the sliding window length is varying. The idea is to check the influence of the length of sliding window on the fusion performance. The values of 2, 5, 10 and 20 are examined. The results are presented in Fig. 4. Fig. 4. Comparison of course estimated for different sliding window length It can be noticed that, the shorter is single window, the more fusion is sensitive for changes. The difference however, except 5 are not so big. It can be assumed that optimal value of sliding window is somewhere between 15 and 20. VI. SUMMARY The paper presented theoretical and empirical research of fusion of target data from tracking radar and AIS in ECDIS. Starting from theoretical aspects including ECDIS, functionalities, radar and AIS characteristic, through concepts of fusion up to practical results of numerical experiment. The empirical research were carried out with the use of real target data recorded at research ship from radar and AIS system. Two different methods of decentralized fusion were examined. It was assumed that, like in real environment only the measurements are known from NMEA strings. Thus covariance matrix has to be estimated. Two ways was proposed. In the first approach covariance matrix was calculated based on IMO accuracy requirements. In the second approach covariance matrix was calculated from variances of measurements in state vectors over a sliding window. The influence of the sliding window length was also examined. The results of the research can be summed up with the conclusion, that parameters of covariance matrix may have an important influence on fusion process. Both examined algorithms of fusion are in fact some kind of weighted average. Thus the weights, deriving from covariance matrix are of vital importance. Applying dynamic values in covariance matrix allows to adjust algorithm better for situation. REFERENCES [1] Stateczny A., AIS and radar data fusion for maritime navigation, Zeszyty Naukowe AM, Szczecin, 2004 [2] Stateczny A. (ed.), Radar navigation, GTN, Gdansk 2011 (in polish) [3] Borkowski P., Data fusion in a navigational decision support system on a sea-going vessel, Polish Maritime Research, Vol. 19, pp.78-85, 2012 [4] Pietrzykowski Z., Borkowski P., Wolejsza P., Marine Integrated Navigational Decision Support System, in Mikulski J.(ed.) Telematics in the transport environment, Book Series: Communications in Computer and Information Science, Vol. 329, pp. 284-292, 2012 [5] Borkowski P., Zwierzewicz Z., Ship course-keeping algorithm basd on knowledge base, Intelligent automation and soft computing, Vol. 17, pp. 149-163, 2011 [6] IMO: Resolution MSC.232(82) - Adoption of the revised performance standards for Electronic Chart Display and Information Systems (ECDIS), 2006 [7] IMO: Resolution MSC.192(79) - Adoption of the revised performance standards for radar equipment, 2004 [8] IEC, IEC 61174: Maritime navigation and radiocommunication equipment and systems Electronic chart display and information system (ECDIS) Operational and performance requirements, methods of testing and required test results, Ed. 3.0, Geneva, 2008. [9] Yang L., Cheng Y., Wei H., Lu J., Error Analysis of Multi-Sensor Data Fusion System for Target Detection on The Ocean Surface, IEEE Inrernational Conference of Information Acquisition, 2006 [10] Kazmierski W., Stateczny A.: Multisensor tracking of marine targets decentralized fusion of Kalman and neural filters. International Journal of Electronics and Telecommunications, 2011 [11] Kazimierski W., Association of targets in the proces of integration of tracking radar and AIS, Scientific Journal of Maritime University of Szczecin Vol. 22(94), pp. 18-22 [12] Kazimierski W., Problems of Data Fusion of Tracking Radar and AIS for the Needs of Integrated Navigation Systems at Sea, IEEE Interational Radar Symposium 2013, in press [13] Bar-Shalom Y. and Li X.-R., Multitarget-multisensor tracking: Principles and techniques. YBS Publishing, 1995. [14] Stateczny, A., Kazimierski, W., A comparison of the target tracking in marine navigational radars by means of GRNN filter and numerical filter. IEEE RADAR CONFERENCE VOLS. 1-4, 2008 Rome, ITALY. [15] Stateczny, A., Kazimierski, W., Selection of GRNN network parameters for the needs of state vector estimation of manoeuvring target in ARPA devices. Photonics Applications in Astronomy, Communications, Industry, and High-Energy Physics Experiments IV Book Series: Proceedings of the society of photo-optical instrumentation engineers (spie) Volume: 6159, 2006. [16] Stateczny A., Kazimierski W.: Determining Manoeuvre Detection Threshold of GRNN Filter in the Process of Tracking in Marine Navigational Radars, Proceedings of International Radar Symposium 2008, Wrocław 2008 [17] Stateczny, A., Neural manoeuvre detection of the tracked target in ARPA systems. CONTROL APPLICATIONS IN MARINE SYSTEMS 2001 (CAMS 2001) Book Series: IFAC. Editor(s): Katebi R. PROCEEDINGS SERIES. [18] Kazimierski W., Stateczny A. Optimization of multiple model neural tracking filter for marine targets. 13 th International Radar Symposium (IRS 2012), 5 th Microwave and Radar Week. Warszawa, 2012, pp. 543-548. [19] Kazimierski W., Łubczonek W. Vertification of marine multiple model neural tracking filter for the needs of shore radar stations. 13th International Radar Symposium (IRS 2012), 5th Microwave and Radar Week. Warszawa, 2012, pp. 554-559. [20] Kazimierski W., Zaniewicz G., Stateczny A. Verification of multiple model neural tracking filter with ship s radar. 13 th International Radar Symposium (IRS 2012), 5 th Microwave and Radar Week. Warszawa, 2012, pp. 549-553.
[21] Kazimierski W. Analysis of tracking errors of multiple model neural filter for marine navigational radar. Elektronika. No 7/2012, ISSN 0033-2089, SIGMA, Warszawa 2012, pp. 47-50. [22] Adamski, M.E.; Kulpa, K.S.; Nalecz, M.; Wojtkiewicz, A., Phase noise in two-dimensional spectrum of video signal in FMCW homodyne radar. Microwaves, Radar and Wireless Communications. 2000. MIKON- 2000. 13th International Conference on, vol.2, no., pp.645-648 vol.2, 2000. [23] Kulpa, K.S. Novel method of decreasing influence of phase noise on FMCW radar. Radar, 2001 CIE International Conference on, pp.319-323, 2001. [24] Kulpa, K.S., Focusing range image in VCO based FMCW radar. Radar Conference, 2003. pp. 235-238, 3-5 Sept. 2003. [25] Kulpa, K.S.; Wojtkiewicz, A.; Nalecz, M.; Misiurewicz, J., The simple method for analysis of nonlinear frequency distortions in FMCW radar. Microwaves, Radar and Wireless Communications. 2000. MIKON- 2000. 13th International Conference on, vol.1, no., pp.235-238 vol.1, 2000. [26] Stateczny A., Methods of comparative plotting of the ship's position. MARITIME ENGINEERING & PORTS III Book Series: WATER STUDIES SERIES Volume: 12, 2002. Editor(s): Brebbia CA; Sciutto G. [27] Stateczny A., Comparative Navigation as an Alternative Positioning System, Proceedings of the 11th IAIN World Congress Smart navigation Systems and Services, 2003, Berlin, Germany. [28] Stateczny, A., Artificial neural networks for comparative navigation. ARTIFICIAL INTELLIGENCE AND SOFT COMPUTING -ICAISC 2004 Book Series: Lecture Notes in Artificial Intelligence Volume: 3070, 2004. Editor(s): Rutkowski L; Siekmann J; Tadeusiewicz R; Zadeh LA. [29] Lubczonek, J., Stateczny, A., Aspects of spatial planning of radar sensor network for inland waterways surveillance. 2009 EUROPEAN RADAR CONFERENCE (EURAD 2009) Book Series: European Radar Conference-EuRAD, 2009, Rome, ITALY. [30] Matzka S., Altendorfer R., A comparison of track-to-track fusion algorithms for automotive sensor fusion, IEEE Inrernational Conference on Multisensor Fusion and Integration for Intelligent Systems, 2008 [31] Liggins M.E (Editor), Llinas J.(Editor), Hall D. L.(Editor): Handbook of Multisensor Data Fusion: Theory and Practice, Second Edition (Electrical Engineering & Applied Signal Processing Series), CRC Press, 2009