Adaptive filtering algorithm in maneuvering target tracking

In our opinion, the commonly accepted 'current' statistical model for tracking maneuvering target suffers from the shortcoming of its heavy dependence on the choice of limit acceleration; this heavy dependence brings frequently the rapid lowering of tracking performance when the real acceleration exceeds the chosen limit acceleration. Our new adaptive filtering (NAF) algorithm bypasses the choice of limit acceleration. Mathematically, the traditional AF (adaptive filtering) algorithm is based on eqs. (1) through (8) in the full paper. Our NAF algorithm is still based on eqs. (1) through (8), but, in addition, it uses eq. (9) to eliminate the need for heavy dependence on the choice of limit acceleration. Eq. (9) is relatively simple and is σ²_a(k) = (2/T²) · |x̂(k|k)-x̂(k|k-1)|, where σ²_a(k), T and x have standard meanings. The physical meaning of eq. (9) is that NAF algorithm expresses variation of acceleration with the information of position to carry out the self adaptation of noise variance. The simulation results indicate that NAF algorithm has a good performance in tracking maneuvering target. The RMSEs (root mean square errors) of position, velocity and acceleration are decreased by about 17%, 10%, and 12% respectively compared with the traditional AF algorithm.