Globally Consistent Mosaicking for Autonomous Visual Navigation

Abstract

Mobile robot localization from large-scale appearance mosaics has been showing increasing promise as a low-cost, high-performance and infrastructure-free solution to vehicle guidance in man-made environments. The feasibility of this technique relies on the construction of a high-resolution mosaic of the vehicle? environment. For reliable position estimation, the mosaic must be locally distortion-free as well as globally consistent. The problem of loop closure in cyclic environments that plagues this process is one that is commonly encountered in all map-building procedures, and its solution is often computationally expensive. This document investigates the problem of map-building with observations having low spatial and temporal persistence from sensors having a short sensory horizon. By exploiting the topology of sensor observations, the problem of mosaicking can be formulated as one of constrained optimization whose solution can be obtained efficiently even for the problem scale typical to the application of interest. A basis of the space of constraints in spatial relationships between observations can be easily extracted. Extrinsically available information in the form of survey data can be treated identically to these constraints and incorporated in the map-building process. The developed framework can also be extended to accommodate incremental construction for online implementation