Learning an object tracker with a random forest and simulated measurements

In this paper, a plain data-driven and simulation-based approach to object tracking is investigated. The basic idea is to use the probabilistic model of the tracking problem to simulate a large amount of state and observation sequences. Both are fed into a regression algorithm that learns a mapping from the observations to the states. In particular, we consider random forest regression and apply it to an object tracking problem using bearing-range measurements. The performance of the random forest tracking is compared to a Kalman smoother and particle filter.