Comparison of Sensor Selection Methods for Markov Localization

Markov localization is one of the effective techniques for determining the physical locations of autonomous objects whose behaviors are nondeterministic and the perceptions of the environment are limited. In a sensor network, the sensor selection problem is concerned with selecting and allocating the available sensors to moving objects in order to obtain the best tracking accuracy for the overall system. In this paper, we formulate the sensor selection problem in a Markov localization setting and examine the condition that ensures the maximal increase of the probability that a target is at its actual location. Based on this condition, we propose a sensor allocation strategy to assign the available sensor resources to individual objects in a grid based surveillance area. We compare our approach with the popularly used mutual information based sensor selection criterion. We found that the proposed method is computationally more efficient and yields more accurate localization results