Monte Carlo Localization in 3D Maps Using Stereo Vision

摘要

In this paper we present a Monte Carlo localization algo-rithm that exploits 3D information obtained by a trinocular stereo camera. First, we obtain a 3D map by estimating the optimal transformations between two consecutive views of the environment through the minimization of an energy function. Then, we use a particle-filter algorithm for addressing the localization in the map. For that purpose we define the likelihood of each sample as depending not only on the compatibility of its 3D perception with that of the observation, but also depending on its compatibility in terms of visual appearance. Our experimental results show the success of the algorithm both in easy and quite ambiguous settings, and they also show the speed-up in convergence when visual appearance is added to depth information.