군함에서의 추적모듈용 α-β-γ 필터와 칼만 필터의 성능비교에 관한 연구

Abstract

Tracking refers to the estimation of the state of a target on motion with some degree of accuracy given at least one measurement. The measurement, which is the output obtained from sensors, contains system errors and errors resulting from the surrounding environment. Tracking filters play the key role of target state estimation after which the tracking system is updated. Therefore, the type of filter used in carrying out the estimations is crucial in determining the integrity and reliability of the updated value. This is especially true since different filters vary in their performance when subjected to different environments and initial conditions of motion dynamics. In addition, applications of different filter design methods have previously confirmed that filtering performance is a tradeoff between error reduction and a good transient response. Therefore, the criteria for selecting a particular filter for use in a tracking application depends on the given performance requirement.

This study explores and investigates the operation of the Kalman filter and three α-β-γ tracking filter models that include Benedict-Bordner also known as the Simpson filter, Gray-Murray model and the fading memory α-β-γ filter. These filters are then compared based on the ability to reduce noise and follow a high dynamic target warship with minimum total lag error. The total lag error is the cumulative residual error computed from the difference between the true and the predicted positions, and the true and estimated positions for the given data samples. The results indicate that, although the Benedict-Bordner model performs poorly compared to the other filters in all aspects of performance comparison, the filter starts off sluggishly at the beginning of the tracking process as indicated by the overshooting on the trajectories, but stabilizes and picks up a good transient response as the tracking duration increases. The Gray-Murray model, on the other hand, demonstrates a better tracking ability as depicted by its higher accuracy and an even better response to a change in the target’s maneuver as compared to the Benedict-Bordner model. The Fading memory model out-performs the other two α-β-γ filters in terms of tracking and estimation error reduction, but based on sensitivity to target maneuvers and variance reduction ratio the Gray-Murray model demonstrates a slightly better performance. The Kalman filter, on the other hand, has a higher tracking accuracy compared to the α-β-γ filters which, however, have a higher sensitivity to target maneuvers and data stability as indicated by the steadier trajectories obtained. These results are a further proof that no one particular filter is perfect in all dimensions of selection criteria but it is rather a compromise that has to be made depending on the requirement of the physical system under consideration.