الفهرس Only 14 pages are availabe for public view
 |  | from | 108 |  |  |
| | | from | 108 | | |
Abstract
Classical target tracking problem has been tackled in numerous research works. However, when the target’s maneuvering is considered, the single Kalman filter approach is unacceptable since the unknown acceleration appears as large process noise in the target model and its variance does not cover it. The first attempt to resolve this problem was made by Singer, who proposed a tracking model in which maneuvering was assumed to be a first-order Markov process with time correlation.
Another interesting solution for this problem is the Interacting Multiple Models (IMM) which has a fixed number of predefined sub-models to describe the motion of target. The first limitation of this model is the fixed number of sub-models which limits tracking new motions that not described in the sub¬models. The second limitation is the predefined models: there is no automated procedure to infer a sub-model in order to describe new observed motions.
In this thesis, an approach to model this problem as a cognitive process is introduced in which motion is captured and stored as knowledge and hence can be recalled or recognized. Two fuzzy-neural based models are suggested in this research; the first one is ”Target Tracking Model” which has the ability to learn new motions as well as tracking maneuvering targets, while the second model is ”Motion Classification Model” which has the ability to classify motion shapes
~) •H~n\)y.