Two Levels Data Fusion Filtering Algorithms of Multimode Compound Seeker

Transcription

1 TELKOMNKA, Vol.11, No.11, November 13, pp. 674~68 e-ssn: 87-78X 674 Two Levels Data Fusion Filtering Algorithms of Multimode Compound Seeker Guodong Zhang*, Zhong Liu, Ke Li College of Electronic Engineering, Naval Universit of Engineering (NUE), 717# Jiefang oad, Wu han, China, Ph./Fa: *Corresponding author, Abstract The multimode compound guidance is an important guidance sstem, which can assurance missile impact accurac in the complicated electromagnetic environment. Aiming at the data fusion filtering problem of multimode compound seeker, this paper presents a two-stage data fusion algorithm. First of all, each sensor data is send into first-stage fusion structure for centralied fusion processing after time registration, and then the processing results are sent into second-stage fusion structure for distributed fusion processing with radar and infrared information processing results. Meanwhile, in order to solve the problems that the different observation information in distributed fusion structure has different dimensions, the observation data is etended to the same dimension. Simulation results show that the proposed algorithm can effectivel improve the data fusion efficienc, and improve the efficienc of the filter convergence accurac and convergence speed, which has a certain guiding significance to engineering practice. Kewords: multimode compound guidance, two-stage data fusion, sub-filter, etended dimension matri Copright 13 Universitas Ahmad Dahlan. All rights reserved. 1. ntroduction The improvement of the guidance abilit and anti-interference abilit has been the focus of research on the accurac guided weapon, so the multimode compound guidance has become an important direction that the accurac guided weapon develops to intelligent weapon. At present, used or developing multimode compound seeker mainl adopts dual mode composite form. adar and infrared sensor compound seeker are the most common form among them. The infrared sensor can measure the angle information of target, which has some advantages such as high angle measuring accurac and continuous measurement, but it cannot get the range information, on the opposite, radar can all-weather measure the target s angle and distance information, but its angle measuring accurac is low, and it will produce missing detective phenomenon in some measure period. Therefore, an effective fusion of their information can improve the guidance s target tracking accurac and enhance its target tracking robustness and anti-interference abilit. Multi-sensor information fusion is the ke technolog in radar and infrared dual mode guidance sstem. eference [1] put forward a radar and infrared collaborative tracking methods, and improved the target tracking accurac b using Unscented Kalman Filter (UKF), reference [] put forward an estimate method using fusion measurement to estimate the target state and reduced the estimation error, reference [3] put forward a method that measurement directl involved in track fusion, which used the public dnamic equation as foundation and got fusion formula, reference [4, 5] discussed sensors track fusion problem which had different measuring accurac, but the sensors data had the same dimension, state matri, measurement matri and state transition matri, reference [6] discussed the data fusion problem about d/3d radar and different d passive sensor, reference [7] discussed asnchronous data fusion problem about multitasking radar and optical sensor. The data fusion methods mentioned above can be divided into centralied fusion and distributed fusion. Centralied fusion is a method that all sensor measurement data are transmitted to a central processor for processing and fusion, and distributed fusion is a method that each sensor data processing intermediate result is transmitted to the center node for processing and fusion. n order to make full use of high accurac angle information of infrared eceived March 17, 13; evised June 14, 13; Accepted Jul 13, 13

2 TELKOMNKA e-ssn: 87-78X 675 seeker, based on the two kinds of fusion, this paper proposes a two-stage data fusion algorithm: first, each sensor data after time registration is send into first-stage fusion structure for centralied fusion processing, and then the processing results are send into second-stage fusion structure for distributed fusion processing with radar and infrared sensor independent processing results. Meanwhile, in order to solve the problems that different observation information in distributed fusion structure has different dimensions, the observation data is etended to the same dimension. n this wa, the data processing module can provide high accurac guidance information to seeker and improve the seeker electronic interference resistance abilit.. The Proposed Method Aiming at the double sensor sstem in a same platform, in order to make effectivel use of the measurement data in fusion processing, the measurement data needs time and space registration before data fusion. The purpose of time registration is to ensure that the data input to the fusion filter is measured at the same time, so that the filter can estimate the accurate target motion state. Because infrared sensor sampling rate is higher than radar sampling rate, so we assume the ratio of radar sampling period T and infrared sensor sampling period T is a positive integer n. f the last track update time ist k 1, then the net track update time is t t 1 nt, so the infrared sensor has n measured values in the interval, t ] [ tk 1 k k k. Using the least square method to fusion the n measured values into a virtual measured value as infrared sensor measurement at the momentt k, which can realie time registration of two sensor measurements. The radar and infrared sensor is distributed in the same seeker platform, so before the data fusion we need to get space registration of data comes from two sensors. The purpose of space registration is to translate the different sensor measurement data into the same coordinate sstem, and then carr on fusion filtering. Assuming the platform coordinate sstem o s origin o is in missile center of o is reclosing with the projectile longitudinal ais and its positive direction is the o is in projectile longitudinal smmetr plane, o ais, which points to the upside is positive, ais o is o, and its direction is determined according to the right-hand mass, ais direction of projectile head, ais perpendicular to the perpendicular to the plane rule. The pitch angle and aimuth angle directed b radar optic ais are and, the pitch angle and aimuth angle directed b infrared sensor optic ais are and, radar installation rolling angle is, infrared sensor installation rolling angle is. Set the declinate between radar and infrared sensor measurement is, and the rectangular coordinate sstem origin of radar in the infrared sensor rectangular coordinate sstem is T E [ e e e ]. T T T are range, pitch angle and aimuth angle of target when the are transformed from radar coordinate sstem to infrared coordinate sstem respectivel. The coordinate transformation process can be completed through the coordinate rotation transformation and coordinates translation transformation: T T T ( ) ( ) ( ) Where: T T cos T sin T T T sin T T T cos T cos T T T T E (1) () Two Levels Data Fusion Filtering Algorithms of Multimode Compound Seeker (Guodong Zhang)

3 676 e-ssn: 87-78X ( ) 1 cos sin sin cos (3) cos sin ( ) 1 (4) sin cos cos sin ( ) sin cos (5) 1, T T T T T T, is the coordinates of target in radar rectangular coordinate sstem, T T, k () is rotation matri when the coordinate sstem rotates degrees around the ais k. as:, is the coordinates of target in infrared sensor rectangular coordinate sstem, Therefore, radar measurement in infrared sensor coordinate sstem can be epressed T T T T T arcsin arcsin T T T T T (6) n order to narrative convenientl, the coordinate transformation of radar measurement T T T are recorded as,. 3. esearch Method When radar measurement data and infrared sensor measurement data are got time and space registration, the will carr on data fusion. Because infrared sensor can onl measure the angle of target information, so in this paper, onl the two kinds of sensor angle information are got fusion. There are man methods for measurement data fusion, and this paper uses the weighted average method for data fusion. According to the principle that fusion measurement information has the minimum root mean square error, we can get the fusion target measurement value. ecording as fusion target pitch angle as fusion target aimuth angle, as fusion pitch angle measurement variance as fusion aimuth angle measurement variance, as infrared sensor pitch angle measurement variance, infrared sensor aimuth angle measurement variance, variance after coordinate conversion, coordinate conversion, then: ( k ) ( ) as as radar pitch angle measurement as radar aimuth angle measurement variance after (7) TELKOMNKA Vol. 11, No. 11, November 13:

4 TELKOMNKA e-ssn: 87-78X 677 ( k) ( ) (8) (9) (1) n order to improve the utiliation rate of infrared data, and get more accurate estimation of first-stage data fusion filtering results, this paper uses distributed filtering thoughts to design three sub-filter. The carr on filtering processing for radar measurement data, infrared sensor measurement data and first-stage fusion measurement data respectivel, then the processing results will be sent to the second-stage fusion filter for state estimation fusion, and the fusion result is used to correct the sub-filter results, which in order to make results have higher accurac than centralied fusion or distributed fusion. adar measurement data are obtained in the polar coordinates sstem, in order to appl it to the standard KF, the measurement data and its mean and variance need to get coordinate conversion. Measurement data in the form of coordinate conversion is such as tpe (3). ecording the state vector as X, the form of converted measurement data mean is as follows: ( / / ) T cos cos ( e e ) ( / / ) cos sin ( e e ) T (11) ( / ) sin ( e e ) T The debased measurement value is: T T T T T T (1) So far we can use the standard KF to filter the measurement. Because infrared sensor can onl observed the target angle data, in order to make it can be applied to kalman filter, we need to use pseudo linear estimation method. n reference [8-1], the pseudo linear estimator is applied in the etended kalman filter to carr on the angle measurement passive tracking. Although this method is a biased estimation, it can overcome the filter divergence phenomenon because it uses the pseudo linear estimator s valuation as kalman filter s initial value. The pseudo linear filter sstem parameters can be epressed as follows: The state vector: X (13) The measurement matri: Z sin cos sin sin cos sin cos (14) Two Levels Data Fusion Filtering Algorithms of Multimode Compound Seeker (Guodong Zhang)

5 678 e-ssn: 87-78X The transfer matri: H cos sin sin sin sin cos cos (15) The measurement data variance: (16) So far we can use the standard KF to filtering the measurement. Due to the measurement matri of each filter has different dimension, in order to make sure that three sub-filter s filtering results can be reprocessing in the same KF filter, dimension of parameters in each sub-filter s measurement equation needs to be etended. Three subfilter s etended measurement matri is as follows: Z sin cos T T sin The transfer matri is: cos cos T sin sin cos sin sin cos cos (17) H 1 cos sin sin sin 1 sin cos 1 cos (18) So far we can use KF for second-stage fusion, the two-stage data fusion algorithm process is as follows: Figure 1. Data Processing Flow Chart of Two Level Fusion Algorithm TELKOMNKA Vol. 11, No. 11, November 13:

6 TELKOMNKA e-ssn: 87-78X esults and Discussion Suppose an anti-ship missile with radar-infrared dual mode seeker uniform flight in a straight line close to sea level at the speed of 1Ma, and its initial position in the geograph rectangular coordinates sstem is ( m,m,m).t s cruise angle is, radar filter s sampling rate is 1 H, infrared filter s sampling rate is 5 H. Our ship is static and the initial position is ( 1m,m,m). The fusion detection data is shown in Figure, three sub-filter s position filtering error is shown in Figure 3, and the contrast diagram of two-stage fusion filter s position filtering results and one-stage fusion filter s position filtering results is shown in Figure 4 and Table 1. m 1 5 ange rad Pitch Angle rad Course Angle Figure. Measurement Data Simulation Diagram As can be seen in Figure 3, three sub-filter s prediction accurac on the measurement data is: fusion sub-filter accurac > infrared sensor sub-filter accurac > radar sub-filter accurac. This phenomenon is determined b the measurement information qualit which input to filter: infrared sensor has high data sampling rate, if it combined with higher accurac initial value, then it can get high accurac filtering results; radar has low data sampling rate, but it is a three coordinates radar, which can provide target range information, so it also can get good filtering results; fusion filter due to combined with both effective measurement information from infrared sensor and radar, so it get the highest accurac filtering results. The filtering result in Figure 4 and Table 1 shows that the two-stage fusion filter has a better performance than onestage fusion filter in convergence accurac and convergence time. The reason is that the twostage fusion filter makes full use of ever sensor s measurement data and the filtering results Two-Stage Fusion Filter One-Stage Fusion Filter 3.5 m Figure 3. Sub-filter s Position Filtering Error Schematic Diagram Figure 4. Contrast Diagram of Fusion Filters Position Filtering Error Two Levels Data Fusion Filtering Algorithms of Multimode Compound Seeker (Guodong Zhang)

7 68 e-ssn: 87-78X Table 1. The Convergence Accurac Comparison of One Stage Fusion Sub-filter and Two Stage Fusion Filter esults one stage fusion sub-filter two stage fusion filter results position convergence precision (m) velocit convergence precision (m/s) position convergence precision (m) velocit convergence precision (m/s) ais ais ais Conclusion Multimode compound guidance is one of the accurac guidance weapon s important development directions, and how to make reasonable and effective use of data information from multisensor is a ver remarkable research direction. n order to improve the multimode compound seeker s target tracking accurac, this paper puts forward a kind of two-stage fusion filter. This filter has a comple structure, and this feature makes it has a higher accurac data processing result. The simulation results indicate that the method can effectivel improve the filtering convergence accurac and convergence time, which has certain engineering guiding significance in improving the multimode compound seeker guidance accurac. eferences [1] Saha K, Chang KC. An efficient algorithm for multi-sensor track fusion. EEE Transactions on Aerospace and Electronics Sstems. 1998; 34(1): -1. [] oecker JA, Mcgillem CD. Comparison of Two-sensor Tracking Methods Based On State Vector Fusion and Measurement Fusion. EEE Transactions on AES. 1988; 4(4): [3] HaiMovich AM, YOSKO J, Greenbeg J, e tal. Fusion of Sensors with Dissimilar Measurement/Tracking Accuracies. EEE Transactions on AES. 1993; 9(1): [4] Castella F. Theoretical Performance of Multisensor Track-to-Track Corrcelation Technique. EEE proceedings of adar, Sonar and Navigation, Washington, D.C. 1995; 14(6): [5] Saha K. Track-to-Track Fusion with Dissimilar Sensors. EEE Transactions on AES. 1996; 3(3): [6] Deb S, Pattipati K, Bar-shalom YA. Multisensor-Multitarget Data Association Algorithm for Heterogeneous Sensors. EEE Transactions on AES. 1993; 9(): [7] Blair WD, ice T, Alouani AT, e tal. Asnchronous Data Fusion For Target Tracking with a Multi- Tasking adar and Optical Sensor. EEE Transactions on Acquisition Tracking and Pointing. 1991; 148(4): [8] Lindgren AG, Gong KF. Position and velocit estimation via bearing observations. EEE Transactions on Aerosp Electron Sstem. 1978; 14(5): [9] Shan-shan Fan, Xuan Yang. 3D Corresponding Control Points Estimation using Mean Shift teration. TELKOMNKA. 1; 1(5): 19-. [1] Qu Jun-suo. A algorithm of fast digital phase modulation signal recognition. TELKOMNKA. 1; 1(8): TELKOMNKA Vol. 11, No. 11, November 13: