Modern Multi-target Tracking And Multi-sensor Integration Of Key Technologies

With the rapid development of electronics, electromagnetics, material, anddynamics, the problem of target tracking has changed significantly, including thesensing environment and target are more complex and changable, the sensing methodsare more various, and the sensing requirements are more ascensive. Some of the keyproblems in the field of modern multi-target tracking and multi-sensor fusion aresystematically studied, such as space alignment of multi-source information, nonlinearestimation, single sensor data association, multi-sensor data association and fusion,ground target tracking, and system performance evaluation. The main contributionsare as follows.1. The typical sampling strategies and location characteristics of Sigma pointsare firstly analyzed according to the conditional function and cost function. Themulti-source information space alignment algorithm based on the UnscentedTransformation (UT) is presented to a certain multi-source information spacealignment problem, and UT methods with different sampling strategies are alsoanalyzed and compared. UT methods are superior to the first-order linearizationapproximation method. The symmetric sampling strategy is the best among samplingstrategies.2. To deal with the multiple modes and nonlinearity in target tracking, theInteracting Multiple Model Unscented Kalman Filter (IMMUKF) is presented basedon the multiple models adaptive estimation method, where the Unscented KalmanFilter (UKF) is used to deal with the nonlinearity of each subfilter. The simulationresults show that our algorithm is better than the traditional IMMEKF algorithm.3. For the multi-target data association problem, the new version of theGeneralized Probabilistic Data Assocation (GPDA) algorithm is proposed. Here weamend the state covariance in the case that the target track has no associatedmeasurement, and introduce the weighted factor to represent the different importantdegree of generalized events based on target or measurement. The simulation resultsshow that the GPDA algorithm is more efficient than the Joint Probabilistic DataAssociation (JPDA). Furthermore, to deal with the maneuvering target dataassociation problem, the combined confirmation gate is constructed to effectivelyintegrate the IMM filter and GPDA, thus the comprehensive Interacting MultipleModel Generalized Probabilistic Data Assocation (C-IMMGPDA) algorithm ispresented. The simulation results show that the C-IMMGPDA algorithm is better thanC-IMMJPDA, and can effectively reduces significant computational burden. 4. The GPDA algorithm is extended from single sensor tracking to multi-sensortracking. For centralized multi-sensor data association, the sequential multi-sensorGPDA (SMSGPDA) algorithm is presented by implementing GPDA in the sequentialstructure. The simulation of muti-sensor data association for RADAR and IRSTshows the efficiency of SMSGPDA. For distributed multi-sensor association andfusion, the rapid track association and fusion algorithm based on GPDA (DMSGPDA)is presented to adapt to the situation that the state convariance of local tracks can notbe achieved. The performance of the DMDGPDA algorithm is analyzed by thesimulation of two RADARs with different measurement accuracy. The results showthat the performance of the DMSGPDA algorithm is better than traditional SimpleWeighted Track Assocation and Fusion algorithm.5. Through extending the hybrid estimation scheme for multi-mode estimationby ultilizing terrain information, the Extended Ground Target Tracking (EGTT)algorithm is proposed, which can effectively reduce the a priori model-set. Based onthe EGTT, The Extended Map-Tuned Variance (EMTV) method is presented, whichincludes the design of a priori model-set and adaptive strategy of model-set. Thesimulation results show that the EMTV is superior to Kalman Filter and Map-TunedVariance (MTV) method for ground target tracking. To deal with the uncertainmaneuver of ground target, the Combined interacting Multiple Model Map-TunedVariance (C-IMMMTV) method is presented. The simulation results show that theC-IMMMTV method can achieve the better performance.6. According to the new requirements of performance evaluation for modernmulti-sensor multi-target tracking in practical application, a set of performanceevaluation indexes defined by the latterest version of national military standard (GJB1904.2-96) are developed and consummated with the detail computation methods insingle sensor system. The new evaluation indexes include track confirmation, trackingaccuracy, stable tracking capability, cross targets tracking capability, group targetstracking capability, processing capacity, processing time delay, and etc. Formulti-sensor system, the performance evaluation indexes of distributed fusion isfurther defined, such as fusion confirmation, time delay in fusion confirmation, fusiontrack stable tracking capability, fusion accuracy, time delay in fusion, and etc. Theevaluation tests for several typical scenarios are implemented based on the systemperformance evaluation platform we developed. The results show that the presentedperformance evaluation indexes can effectively represent the performace of trackingalgorithms, and provide strong support of the improvement of algorithms anddevelopment of systems.