Visual sonar based localization using particle attraction and scattering

A novel real time algorithm for incremental localization is presented based on particle filtering and omnidirectional (360 degree) reconstruction of the environment in visual sonar mode. This omni-directional field of view is economically obtained by temporally accumulating forward views of a single camera which lends itself to alleviation of the perceptual aliasing problem. This algorithm basically draws upon the sampling concept of Monte Carlo localization and the particle attraction concept. However, a new concept of particle merge and scattering creation has been developed which substantially speeds up the localization process. This algorithm has been applied to a typical cluttered office environment for a real robot and tested over many hours of runs with satisfactory performance.