Hide/Show Apps

Performance metrics for fundamental estimation filters

Akçay, Koray
This thesis analyzes fundamental estimation filters ا Alpha-Beta Filter, Alpha-Beta-Gamma Filter, Constant Velocity (CV) Kalman Filter, Constant Acceleration (CA) Kalman Filter, Extended Kalman Filter, 2-model Interacting Multiple Model (IMM) Filter and 3-model IMM with respect to their resource requirements and performance. In resource requirement part, fundamental estimation filters are compared according to their CPU usage, memory needs and complexity. The best fundamental estimation filter which needs very low resources is the Alpha-Beta-Filter. In performance evaluation part of this thesis, performance metrics used are: Root-Mean-Square Error (RMSE), Average Euclidean Error (AEE), Geometric Average Error (GAE) and normalized form of these. The normalized form of performance metrics makes measure of error independent of range and the length of trajectory. Fundamental estimation filters and performance metrics are implemented in MATLAB. MONTE CARLO simulation method and 6 different air trajectories are used for testing. Test results show that performance of fundamental estimation filters varies according to trajectory and target dynamics used in constructing the filter. Consequently, filter performance is application-dependent. Therefore, before choosing an estimation filter, most probable target dynamics, hardware resources and acceptable error level should be investigated. An estimation filter which matches these requirements will be ءthe best estimation filter̕.