MULTIPLE TARGET TRACKING IN NON-GAUSSIAN NOISE (ESTIMATION, SIMULATION TECHNIQUE)

The problems of multiple targets tracking can be divided into two areas: (i) uncertainty of association, (ii) filtering in non-Gaussian noise. The first problem deals with the situation that given one set of the received observations, it is difficult to find which data are associated with targets. The situations such as missing target, false observation, initiation, split, merging, annihilation, maneuver are often found in the real environment and cause the data association problem to be the vital part in the tracking.; Most tracking algorithms in the literature track a fixed number of targets. Those that can handle initiation and annihilation use some assumptions which are either unrealistic or need some parameters which are hard to obtain. A new algorithm is proposed which handles all the situations we discuss above and takes a more realistic approach. Simulations have been done to compare the new algorithm with the nearest neighbor and Bar-Shalom's method when the number of targets is fixed. The ability to handle the situation such as initiation, split, merging, annihilation, maneuver was also tested. The superior results were shown in the simulations for both cases.; For the second problem, almost all the tracking algorithms assume that the observation noise is Gaussian, and use a Kalman filter to do the filtering. In the real environment, it is found that the observation noise is heavy-tailed non-Gaussian. This dissertation attacks this non-Gaussian noise problem by two approaches: (i) Gaussian sum filter, (ii) robust filter. The Gaussian sum filter is found not suitable to use, and the robust filter is adopted. The robust filters were studied by using our extensive Monte Carlo simulation. Various robust filters were compared under different situations and a recommendation was given to choose the suitable robust filter. Also two simulation methods were proposed to further probe the robust filters in the future.; Finally, a robust filter was used in the tracking system under heavy-tailed non-Gaussian noise. The Monte Carlo simulations were conducted to test the combination of the robust filter and the new algorithm and gave encouraging results.