Comparing the state estimates of a Kalman filter to a perfect IMM against a maneuvering target
Mark Silbert, Shahram Sarkani, Thomas A. Mazzuchi · International Conference on Information Fusion · 2011
Tracking maneuvering targets is an important problem. A study was previously performed to compare the state estimation accuracy of a Kalman filter to an interacting multiple model (IMM) for a maneuvering target. The authors defined a maneuvering index to quantify the degree of maneuvering. Their study then compared the state estimates of the two filters as a function of this index. Their results showed that an IMM provides significant improvement over a Kalman filter. That study was revisited and this paper discusses the differing results observed. Our results show that the IMM does improve overall state estimations but much less than in the previous study. This improvement is due to the smaller state estimation errors that the IMM provides over the Kalman filter during the non-maneuvering intervals, rather than the complete domination in performance of the IMM that the previous study revealed. As a result, the 0.5 rule that the previous authors identified, should be revised.