Improved Spectra-based Map Merging using Visual Object for Multi-Robot Systems

Abstract

This paper proposes the improved spectra-based map merging method using geometrical information extracted from visual objects. The system performs exploration by fusing vision information and laser scan matching. The combination of vision and laser range finding is effective and it makes the performance of the system more robust because they reinforce one another in the environment of sufficient structure and visual information. In our work, each member of a robot team performs laser scan matching to build an individual grid map and recognizes objects using a camera. Geometrical information is extracted from the recognized objects and used in the spectra-based map merging (SMM) algorithm. The geometrical information consists of surface normal vectors and position vectors of the objects. The search space for the SMM is greatly reduced due to the geometrical information which works as a strong landmark. The reduced search space solves the problem of multiple local maxima in the SMM algorithm. Also, the minimum overlapping areas of the maps required to perform the SMM is greatly reduced because we can use the positional information of the objects. Consequently, the overall performance of the system becomes more efficient and precise. Experimental results show that the proposed system successfully performs the map merging with relativelysmall overlapping areas between the maps.