Research On Target Tracking Algorithm

Target tracking technology has a wide range of applications in national defense and civil fields and is the key technology of realizing information fusion. Distributed data fusion becomes one of the most important topics in control and estimation area, because it gains advantages over centralized fusion system. But the distributed fusion system has some flaws that do not appear in centralized fusion system—data redundancy. Data redundancy causes data correlation between the fused variables and usually the degree of correlation between variables is not available. In this case, the covariance intersection algorithm (CI) is applicable. CI algorithm can gain a consistent estimation, when the correlation between the fused variables is unknown.Multi-target tracking is another focus in target tracking and estimation area. In multi-target tracking, the traditional algorithms solve the problem through probability data association, which concerns about target-measure association and uses single target tracking algorithm to estimate target state separately. With the increasing of target number and measurements, the cost of computing probability data association will increase explosively. The probability hypothesis density (PHD) filter avoids target-measure association problem. At present, many scholars research tracking method based on PHD filter and its applications. This dissertation gives some research works based on the tracking algorithm referred , and the main work is described as follows:1. In this dissertation, some traditional target tracking algorithms and their basic theory are discussed and analyzed, especially some single target tracking algorithm and multi-target tracking algorithm. Here we combine covariance intersection (CI)algorithm with probability data association(PDA)filter, and realize the application of CI algorithm in distributed system in single target tracking with clutter. And the simple convex combine data fusion algorithm(CC) is discussed and realized in the same distributed system. Through comparing the simulation results of CI algorithm and CC algorithm, we can see the effectiveness of CI algorithm in dealing with correlated data fusion problem.2. This dissertation proposed an improved CI algorithm based on the geometrical description of the CI algorithm. The calculation of optimal constraint value of CI algorithm is time consuming. To reduce the time cost of CI algorithm , a sub-optimal calculation method of constraint value is proposed. The improved method can significantly reduce the time cost and obtain acceptable data fusion accuracy. The improved CI algorithm is realized in distributed system and the result is analyzed and compared with that of CI algorithm to show the effectiveness of the improved method.3. Researches on multi-target tracking problem with unknown target number. The application of probability hypothesis density (PHD) filter is introduced. The application of particle PHD filter (P-PHD)is realized in multi-target tracking model with random target number. The simulation result shows that the P-PHD method is efficient in estimating the target number and their states.4. This dissertation proposed an improved method based on P-PHD filter. The main idea of the improved P-PHD filter is to mark the particles that represent the same target with a unique label, thus the target trace can be obtained when the target state is extracted from a set of particles. And to some extent the improved method can improve the accurate of target number estimation and target state estimation.