Adaptive mode transition matrix for variable sampling time

A conventional dynamic target tracking algorithm can be divided into two steps. A previous estimate is propagated to current scan by using pre-designed dynamic model in the prediction step and the predicted estimate is corrected by using the current measurement in the update step sequentially. To reduce the prediction errors, one has to carefully choose the dynamic model carefully. However, exact target dynamic model is rarely known a priori. Multiple models filters which use several dynamic models present a solution. Interacting Multiple Models (IMM) filter calculates the state estimates and model probabilities for each model. In the interacting step of the IMM filter, the model estimates are mixed using the model transition matrix and the model probabilities. The elements of the model transition matrix are usually constant. This can cause performance degradation for target tracking with variable sampling periods or asynchronous multi sensor systems. We propose an adaptive model transition matrix which adjusts its elements based on variable time interval.