This paper analyses the performance of Kalman filter based algorithms for tracking of a moving person observed by UWB sensors in an indoor environment. It is shown that known tracking algorithms cannot correctly cope with sudden changes - maneuvers - in the movement of the localized person. The article proposes a new algorithm. It combines the input selection approach, which treats maneuvers as non-random variables, with the maneuver detection that is known e.g. in the approach with adjustable noise level. A simulated example compares performance of selected tracking algorithms. It demonstrates that the proposed algorithm outperforms known tracking algorithms.